Poverty Risks after Relationship Dissolution and the Role of Children: A Contemporary Longitudinal Analysis of Seven OECD Countries

Abstract

The divorce literature has consistently found that—especially women—are negatively affected by relationship dissolution in terms of material wellbeing. There is, however, considerable debate on whether these effects are persistent or temporary. We use fixed effects models and control for the socioeconomic status of individuals who separated between 2011 and 2018 in seven countries for which large scale longitudinal data has recently been harmonized in the Comparative Panel File. We find that the transitory nature of the effect of relationship dissolution on poverty risks for women is similar across countries, but also for some men. We further focus on the role of children in the immediate changes in poverty risks after separation, and again find significant differences between countries. We discuss these findings in light of social policies adopted by these countries, more specifically child and spousal support schemes. We find no distinguishable differences in these support schemes that adequately explain the observed dissimilarities. The implications of this study for the future study of the association between relationship dissolution and poverty are discussed and future pathways are suggested.

1. Introduction

The study of poverty risks traditionally considers two pathways into poverty. On the one hand, poverty is considered to be associated with social stratification. These studies focus on characteristics of the individual, such as social class, ethnicity, gender, or education (Breen 2005; Shildrick and Rucell 2015; Townsend 1954). These studies typically understand poverty risks as being associated with long-lasting, almost inescapable features of the individual. On the other hand, life course studies have linked poverty risks to transitions, trajectories, or events (Dewilde 2003). Examples of these occurrences in a person’s biography include childbirth, relationship dissolution, unemployment, or leaving the parental home. The nature of the association with poverty in the life course perspective is a transient one. The risk of becoming poor is temporarily increased after these so called biographical breaks (Elder 1995; Popova and Navicke 2019; Vandecasteele 2010a). More recent research has shown that combining the two perspectives can be a fruitful research strategy, as neither the risk of certain life course events nor the magnitude and duration of the consequences appear to be randomly distributed across individuals or social groups (Härkönen and Dronkers 2006; Oris et al. 2017; Vandecasteele 2010a; Whelan and Maître 2008).

Whether it be from enduring social stratification, transient life course events, or a combination of the two, social policy has been shown to play an important role in mitigating poverty risks (Kangas and Palme 2000; Korpi and Palme 1998; Popova and Navicke 2019). This has been carried out through taxation schemes, systems of benefits, or regulations concerning, for instance, child or spousal support. Particularly productive has been comparative research originating with the seminal work of Esping-Andersen (1990) on different types of welfare states (DiPrete 2002; Kauppinen et al. 2014; Vandecasteele 2010b).

The effects of relationship dissolution are not restricted to the former partners. A better understanding of how poverty is associated with relationship dissolution and comparative research that uncovers differences in these risks is, therefore, necessary not only for the wellbeing of expartners but also to increase the life chances of children growing up with divorced parents. In this regard, especially, mothers and children living with lone mothers have been shown to be particularly vulnerable (Amato 2000, 2001, 2014; Mortelmans 2020; Van Lancker and Vinck 2019). In the present study, we aim to contribute to this body of research by looking at poverty risks following the dissolution of cohabiting heterosexual relationships (legal cohabitations as well as marriages) in seven OECD countries.

Our research adds to the existing literature in several ways. First, one of the unfortunate second order effects of a highly influential framework such as the typology proposed by Esping-Andersen (1990) is that, for a long time, comparative research tended to be confined to those countries that more or less fit into categories that were used to describe European countries. In the present study, we include two countries that have, thus far, been underrepresented in comparative research of this kind: Russia and South Korea. By using harmonized data from seven of the longest running panel studies, harmonized by the Comparative Panel File project (Turek et al. 2021), we are able to offer an initial bird’s eye view of the impact of relationship dissolution on poverty risks for men and women in Australia, South Korea, the United States, Russia, Switzerland, Germany, and the United Kingdom. Next, by looking at dissolutions of cohabitational relationships as well as marriages, we capture what Van den Berg and Mortelmans (2018) have coined the “invisible divorce wave”. Third, by using longitudinal data, we are able to statistically control for the effects of fixed characteristics. This means that we get clearer estimates of the transitory effect of relationship dissolution itself on poverty risks, decreasing the bias that results from inadequately distinguishing between life course events and social stratification.

2. Theoretical Background: Poverty, Relationship Dissolution, and Social Policy

There is no single definition of what is meant by being poor or, at the aggregate level, what poverty stands for. Hagenaars and De Vos (1988) asserted that, at the time, all definitions fall within one of three categories of being either absolute, relative, or subjective. They go on to show that different definitions can lead to wildly differing estimates of the proportion of the population that is poor. Two of the questions that have arisen in the poverty literature centre around (a) its durability over time and (b) how to measure its level at a certain point in time.

Rodgers and Rodgers (1993) argue that the choice of time period is central to the measurement of poverty. They distinguish between two types of poverty over time. The first is persistent—or chronic—poverty. People experiencing persistent poverty do so over a longer period of time. This often includes the transmission of poverty across generations. The second type of poverty is labelled transitory poverty and concerns poverty stemming from fluctuations in a household’s welfare (McCulloch and Baulch 2000), often following certain life course events such as the death of a family member, the birth of a child, job loss, or divorce (Dewilde 2003).

The debate on how poverty should be measured is far from over and new approaches appear to be suggested seasonally (considering the continuing academic interest in the field, this quip might just as well be true). Traditional income based measurements have famously received criticism for being unidimensional and failing to take into account other aspects of the quality of life (Nussbaum 2006; Sen 2006). Multidimensional attempts at measuring poverty such as the Human Development Index (Anand and Sen 1994) have, in turn, received criticism for being too simplistic and not supported by theory, resulting in an “industry of multidimensional index building” (Grusky and Weeden 2013; Sagar and Najam 1998). More recent proposals attempt to take gendered experiences of poverty into account (Bessell 2015; Pogge and Wisor 2016).

Unfortunately, there is no clear cut solution to this problem and researchers are left with the “choosing is losing” dilemma. The reason for this is rather obvious: poverty is not a self-defining concept (Bellu and Liberati 2005). In lieu of adding to the debate, we resign ourselves to a commonly used measure of poverty, the at risk of poverty (AROP) rate. This measure (used by Eurostat) defines anyone falling below the threshold of 60% of the national median of equivalized household income as at risk of being poor. While this measure is met with as much critique in the scientific literature as any other (the most evident flaw being that by using this relative measure it is by definition impossible to eradicate the risk of poverty), it is perfectly suited for comparative research focussed mainly on uncovering differences between countries in its association with relationship dissolution.

On that subject, much research has been carried out on the consequences of divorce, or relationship dissolution more generally, for expartners and children. Taken as a whole, the event of relationship dissolution is consistently shown to have negative consequences for both expartners as well as their children. The two main reasons for this detrimental effect of divorce on poverty risks are the loss of one partner’s income, which is exacerbated by the loss of economies of scale (Couch et al. 2013). Reviews show that these consequences consist of declines in material, physical, and mental wellbeing (Amato 2000, 2014; Mortelmans 2020). A consistent finding in the literature is that women suffer the most financially after divorce (Andreß et al. 2006; Bayaz-Ozturk et al. 2018; De Vaus et al. 2017; Hauser et al. 2018) or after separation of a cohabiting partnership (Avellar and Smock 2005; Manting and Bouman 2006; Tach and Eads 2015). Although there are substantial cross-national differences, men are shown to remain stable or even improve their economic situation after the dissolution of a marriage, McManus and DiPrete (2001) being a notable exception to this finding. This is often attributed to child custody more often being awarded to the mother and the stronger labour market attachment of men during the marriage (Mortelmans 2020). More specifically, on the association with poverty, Leopold (2018) found that, in Germany, women’s poverty risks surged for a short period after divorce while men’s remained stable. Similar results were found in Canada (Gadalla 2008) and the Netherlands (Hogendoorn et al. 2020).

Considering the host of undesirable outcomes that are associated with poverty, it is no surprise that most governments implement social policies in an attempt to reduce it. Moreover, the importance of public policies in mitigating the negative consequences of certain life course risks, i.e., those associated most often with decreased wellbeing and lower living standards, are of particular interest to comparative researchers (DiPrete 2002). In the case of relationship dissolution, the effects on poverty risks are most likely affected by policies and regulations relating to either spousal maintenance or child maintenance.

In general, spousal support in any of the countries under observation in this paper (except Korea) is only mandated in specific circumstances, such as old age and disability, but also in the case of unequal earning power after a long marriage. In Germany, for instance, alimony reforms were made in 2008 with the explicit purpose of emphasizing post-marital self-responsibility (Bredtmann and Vonnahme 2017), but equity is still one of the grounds for awarding maintenance. While Korea has no system of spousal maintenance, it appears that, in practice, upon divorce, expartners are invariably entitled to receive half of the other party’s assets. This means that, while spousal maintenance might be low or non-existent (De Vaus et al. 2017), this might be compensated through property division.

Table 1. Main characteristics of spousal support systems after relationship dissolution in the selected countries.

Country	Incidence	Grounds c	Determined by	Amount	Duration	Exists for Cohabitation
Australia	Uncommon	Limited circumstances	Court	Minimal	Usually a short period of time	No
Korea	Non-existent	N/A	N/A	N/A	N/A	N/A
USA b	Common	Limited circumstances	Partners, Mediator, Court	Individually determined	Set date or certain life course events a	No
Russia	Uncommon	Limited circumstances	Court	Sufficient for living needs	Discretion of the court	No
Switzerland	Uncommon	Limited circumstances	Court	Sufficient for living needs	Usually limited duration	No
Germany	Common	Limited circumstances	Court	Individually determined	Individually determined	No
UK	Common	Limited circumstances	Court	Sufficient for living needs	Depends on marriage duration and events (remarriage)	Yes

Notes: ^a remarriage (or cohabitation), considerable increases in receiver’s earnings, …; ^b There is considerable variation in rules and regulations across individual states in the US. ^c “Limited circumstances” include, for example: old age, disability, or a prolonged withdrawal from the labour force (i.e., homemakers). Sources: (Bredtmann and Vonnahme 2017; De Vaus et al. 2017; Fehlberg 2004). For Russia: Thomson Reuters Practical Law: “Family law in the Russian Federation: overview”, http://shorturl.at/qtRU9, accessed on 19 January 2022.

shows that the characteristics of spousal maintenance in all seven countries are more or less similar.

All countries in this study require child support payments after parental divorce. As can be seen in

Table 2. Main characteristics of child support systems after relationship dissolution in the selected countries.

Country	Involvement in the Determination of Child Maintenance			Responsibility for Determining Maintenance Payments	Rules for Determining the Amount of Payments	Enforcement of Payments	Different for Children of Unmarried Parents	Age at Which Support Ends	Advance on Maintenance Payments
	Parents	Court	Agency
Australia	Yes—formal system if parents cannot agree	Yes (residual role)	Yes—CSA	Parents or CSA	Rules/rigid formula	CSA	No	18 years or end of schooling	No
Korea	Yes—ratified by court	Yes	No	Parents or Court if parental disagreement	Mostly discretion, no fixed rules or methods	Court	No	Parental agreement or 20 years	No
USA	Yes—ratified by court	Yes	Yes—CSA (varies by state)	Court	Formal guidelines	Court and CSA	No	Varies across states (16–25)	No
Russia	Yes—formal system if parents cannot agree	Yes (residual role)	No	Parents or court if parental disagreement	Rules in case of disagreement	Court	No	18 years	Yes
Switzerland	Yes—ratified by court	Yes	No	Parents with supervision of lawyers or court	Rules	Court	Yes	18 years or end of education	Yes
Germany	Yes	Yes	No	Parents or court if parental disagreement	Mostly discretion, using “support tables”	Court	Yes	18 years	Yes
UK	Yes—ratified by court	Yes (residual role)	Yes—CSA	Parents or CSA	Rules/rigid formula	Court and CSA	No	16 years or 19 years if in full time education	Yes

Notes: CSA = Child Support Agency; PWC = parent with care; there is considerable variation in rules and regulations across individual states in the US. Source: OECD family database (https://www.oecd.org/els/family/database.htm) accessed on 19 January 2022. Last updated: 2010. For Russia: Thomson Reuters Practical Law: “Family law in the Russian Federation: overview”, http://shorturl.at/qtRU9, accessed on 19 January 2022.

, contrary to spousal maintenance payments, the countries in our study show differences between, for instance, how the amount is decided (and who decides it) or the duration of the payments. Considering the transitory nature of the association of divorce as a life course event and poverty, the existence of advances on alimony payments, which might be received before the divorce is legally finalized, could possibly play a role in mitigating the negative economic consequences. Regardless of institutional context, it is important to note that, both for child maintenance and spousal support, it is overwhelmingly the male expartner who pays and the woman who receives the support.

Since poverty, in this study, is measured based on equivalised household income, it seems almost tautological that physical custody arrangements play an important role. The theoretical grounds are straightforward: expenses are higher with children living in the household. While recent data on child custody arrangements are not always available, there are indications that custody arrangements vary greatly between these countries. In Korea, joint custody is virtually non-existent (<1%) and children most often reside with either their mother (56%) or their father (44%) (Chung and Emery 2010; Park 2015). In the United States, there has been a shift towards favouring some form of shared custody (either equal or unequal). In 2010, mother’s sole custody was most common, making 42% of all custody arrangements, followed by equal joint custody (35%), unequal joint custody (15%), father’s sole custody (6%) and split custody, where one child has a different arrangement from their siblings(s) (<1%) (Meyer et al. 2017). Data for the UK are sparse, as 90% of separating couples make private arrangements outside of the court system. Joint custody is estimated to be between 3–17% (Haux et al. 2017). Data on the exact prevalence of father’s versus mother’s sole custody are, to the best of our knowledge, not available, but mother’s sole custody makes up the bulk of the remaining arrangements. Similarly, for Germany, shared physical custody arrangements are estimated to be between 4 and 12%; again, the other arrangements are usually sole custody awarded to the mother, although there are differences in the frequency of paternal contact (Walper et al. 2021). In Russia, child custody is awarded to mothers in more than 90% of the cases (Khazova 2005; Maes et al. 2020). For Australia, a report from 2010 found that only 8% of arrangements were shared custody (Cashmore et al. 2010). Again, the overwhelming majority of other cases are sole custody awarded to the mother. Finally, in Switzerland, one study found that mother’s sole custody made up 61% of cases, versus 34% shared custody and 5% father’s sole custody.

In the following we will analyse the changes in the poverty risks associated with the event of relationship dissolution in the seven countries discussed above. On the one hand, the theoretical and empirical work on poverty over the life course allows us to make some specific hypotheses on the trajectory of poverty risks surrounding separation. On the other hand, the overviews of spousal and child support schemes do not result in the countries under observation falling into some number of categories. There are many similarities and the differences do not fall along some fault lines. As a result, we will formulate some general non-directional hypotheses on country differences.

Presented here, we test the following hypotheses:

Hypothesis 1.

On the transitory effect of relationship dissolution on poverty risks.

Hypothesis 1a.

Across countries, there is an immediate increase in the probability of being at risk of poverty (AROP) after the factual separation of previously married or cohabiting heterosexual couples.

Hypothesis 1b.

There is a decrease over time after the factual separation in the probability of being AROP.

Hypothesis 2.

On the gendered nature of this effect.

The positive effect of relationship dissolution on the probability of being AROP is larger for women than for men.

Hypothesis 3.

On the comparison between countries.

The positive effect of relationship dissolution on poverty risks differs across countries.

Hypothesis 4.

On the role of children.

The effects of the presence of children in the household on poverty risks after relationship dissolution differs across countries.

3. Data and Methods

We use data from long running panel surveys of 7 different countries: (1) The Household, Income and Labour Dynamics in Australia (HILDA); (2) the German Socio-Economic Panel (SOEP); (3) the British Household Panel Surveys (BHPS) and the UK Household Longitudinal Study (UKHLS); (4) the Korean Labour and Income Panel Study (KLIPS); (5) The Russia Longitudinal Monitoring Survey (RLMS); (6) The Swiss Household Panel (SHP); and (7) the United States’ Panel Study of Income Dynamics (PSID). Recently, Turek et al. (2021) have harmonized a subset of the information in these 7 panels in the Comparative Panel File (CPF). This harmonization provides us with a unique opportunity to compare poverty trajectories surrounding relationship dissolution across countries in arguably a more uniform manner than previous studies. While a complete description of both CPF and the separate panel studies is too lengthy to include in this paper, we refer to the CPF’s data manual for an extensive overview (Turek et al. 2020).

From each of these panel files we included individuals between the ages of 18 and 65 for whom the data shows a first divorce or dissolution of a cohabiting heterosexual relationship between 2010 and 2018. Subjects who already entered the sample with a partner status of “divorced” or “separated” were dropped. The resulting individuals were observed starting 10 years prior to the factual separation (the earliest observation being 2001) and a maximum of 8 years afterwards. Since not all separations occurred in the same year, not all individuals could be observed for the same amount of time, meaning that some enter the sample closer to the separation or leave the sample faster afterwards. This resulted in 6034 observation points for a sample of 898 men and 10,422 observations for 1595 women across the 7 countries.

3.1. Dependent Variable

The risk of poverty is measured by first taking the net adjusted disposable household income after taxes and transfers of the gross sample of the CPF harmonized dataset (2,099,534 observations for 318,016 respondents). This income is equivalized by dividing it by the square root of the number of household members. Then, the median for each year in each country was calculated. Finally, the AROP indicator is constructed as a dummy variable with the value of 1 for everyone for whom equivalized income was below 60% of the median income of that country for that year.

3.2. Independent Variables

Time

Time (T) is measured as observation waves. In general, these are around one year apart, except for the U.S. data from the PSID, which has biannual information. Factual separation occurs somewhere between T = −1 and T = 0. An additional time variable is entered into the equation as a dummy variable taking on the value of 0 prior to the separation and 1 afterwards. When interacted, these variables allow us to observe both the trajectory prior to separation, the immediate difference at the time of separation, and the change in trajectory after separation.

3.3. Covariates

A host of covariates were introduced into the models. The first being the country of residence. Next, education was measured as a categorical variable with 3 categories based on the ISCED-97 or ISCED-2011 classification as follows: lower education (ISCED 0–2), medium education (ISCED 3–4), and higher education (ISCED 5–8). We also include the socioeconomic index of the current occupational status based on ISCO version 2008 (ISEI-08). This indicator was constructed according to Ganzeboom (2010) algorithms (Turek et al. 2020). Number of hours worked per week was available for all countries and included as a continuous variable. Two observations were culled due to them being impossibly or improbably high work hours (over 150 h per week). Age was recoded into 5 categories: 18–25, 26–35, 36–45, 46–55, 56–65. Finally, we included a dummy indicator for whether or not there were children living in the household. Due to different measurements by country, this indicator represents the number of children living in the household up to the age of 15 in the UK and 17 years old in all other countries.

Table 3. Descriptive statistics of people experiencing relationship dissolution after 2010 in 7 OECD countries, divided by gender.

	Women (Obs. = 10,422; N = 1595)				Men (Obs. = 6034; N = 898)
Variable	Mean	Std. Dev.	Min	Max	Mean	Std. Dev.	Min	Max
Time to/since factual separation	−1.32	3.94	−10	7	−0.99	3.88	−10	7
ISEI08 b	40.37	14.76	10	89	39.66	13.96	10	89
Weekly work hours	31.80	14.61	0	120	42.15	14.04	0	119
	%				%
AROP c	33%				31%
Country
Australia	20%				31%
Korea	5%				8%
USA	8%				9%
Russia	14%				11%
Switzerland	10%				8%
Germany	10%				7%
UK	33%				26%
Educational attainment
Low	16%				21%
Medium	54%				58%
High	30%				21%
Age
18–25	13%				13%
26–35	31%				31%
36–45	33%				29%
46–55	19%				21%
56–65	4%				6%
Children living in the household a	64%				51%

Notes: ^a Children aged up to 15 years old in the UK, up to 17 years old in other countries. ^b ISEI08 is a socioeconomic index of the occupational status. ^c AROP = at risk of poverty, i.e., an equivalized household income lower than 60% of the median income of that country in that year.

shows the descriptive statistics separated by gender. Descriptive statistics separated by gender and/or country are available from the authors upon request.

3.4. The Models

The evolution of the probability of being AROP surrounding factual separation and the effects of the covariates are estimated through logistic panel regression with individual fixed effects (see: Allison (2009) for a detailed discussion of these models). By doing this, we can eliminate bias due to unobserved heterogeneity as long as these unobserved characteristics are time invariant. In our models, if we assume that poverty risks due to social stratification are made up of a time invariant, inherent component and a time variant component, using fixed effects regression will then control for that part of an individual’s social stratification based poverty risks that are constant over time. This means that—depending on how much social stratification is actually fixed—we can obtain more accurate estimates of the effect of relationship dissolution on the poverty risk. By using interactions between country and time, we circumvent the limitations of the fixed effects model while assuming that there are no differences between countries in terms of the covariates that are entered into the model, but only differences in the overall trajectories.

Figure 1 illustrates a theoretical scenario the model would predict if there was relative stability in the AROP prior to the separation, then a sudden increase immediately after separation, which subsequently decreases over time back to some new (possibly) stable trajectory. The immediate effect of the separation on the probability of being at risk of poverty is then the difference between point A’ (the estimated probability) and point A (the estimated probability based on the pre-separation trajectory). Algebraically, this comes down to a change in intercept that is captured by the dummy variable taking on the value of 1 after separation. We will therefore refer to it as the “intercept effect”. The trajectory of AROP starting from A’ is then captured by combining the baseline estimate of the time variable and the interaction of these polynomials with the dummy variable indicating the time after separation. As the quadratic function estimates the trajectory of the AROP indicator over time, we will refer to this change as the “trajectory effect”.

4. Results

The nature of these models of change, especially when comparing several countries, is such that tables quickly become unwieldy. In our case, this is exacerbated by the more difficult interpretations of odds ratios that are obtained from any logistic regression, let alone when using two- and three-way interactions. We have, therefore, moved the large

Table A1. Results of logistic regressions with fixed effects on the poverty risks surrounding the dissolution of a cohabiting relationship. Results presented as odds ratios.

Model	(1)	(2)	(3)		(4)
	Pooled	Pooled	Women	Men	Women	Men
Prior to separation
Slope	0.992	0.975	0.928	0.792 **	0.951	0.785 **
	(0.013)	(0.018)	(0.055)	(0.065)	(0.057)	(0.065)
Slope x Women		1.019
		(0.023)
Slope x Country
Australia			1.072	1.273 **	1.042	1.278 **
			(0.073)	(0.111)	(0.072)	(0.112)
Korea			1.021	1.377 **	0.949	1.331 **
			(0.079)	(0.135)	(0.076)	(0.133)
USA			1.125	1.274 *	1.097	1.282 *
			(0.081)	(0.122)	(0.080)	(0.124)
Russia			1.017	1.175	0.989	1.190
			(0.070)	(0.112)	(0.070)	(0.114)
Switzerland			1.045	1.092	0.999	1.117
			(0.079)	(0.105)	(0.078)	(0.110)
UK			1.147 *	1.285 **	1.123	1.301 **
			(0.076)	(0.115)	(0.076)	(0.117)
Children in HH					0.219 **	1.484
					(0.115)	(1.055)
Children in HH x Country
Australia					4.007 *	0.836
					(2.346)	(0.625)
Korea					2.103	0.409
					(1.437)	(0.352)
USA					4.117 *	1.264
					(2.667)	(1.041)
Russia					4.942 **	1.118
					(2.936)	(0.884)
Switzerland					3.993 *	9.996 **
					(2.617)	(8.878)
UK					5.659 **	1.552
					(3.218)	(1.167)
Postseparation
Intercept Change	3.229 ***	1.144	17.343 ***	3.890 ***	4.931 ***	3.987 *
	(0.205)	(0.118)	(5.016)	(1.571)	(2.250)	(2.751)
Slope Change	0.912 ***	0.985	0.634 ***	0.919	0.611 ***	0.926
	(0.018)	(0.032)	(0.059)	(0.138)	(0.058)	(0.139)
Intercept Change x Women		5.069 ***
		(0.673)
Slope Change x Women		0.900 **
		(0.036)
Intercept Change x Country
Australia			0.349 **	0.313 **	1.169	0.259
			(0.124)	(0.139)	(0.604)	(0.188)
Korea			0.187 ***	0.176 **	0.263 *	0.133 *
			(0.089)	(0.097)	(0.167)	(0.108)
USA			0.132 ***	0.125 ***	0.345	0.111 **
			(0.052)	(0.066)	(0.207)	(0.093)
Russia			0.189 ***	0.224 **	0.525	0.260
			(0.067)	(0.111)	(0.282)	(0.200)
Switzerland			0.690	0.133 ***	1.342	0.287
			(0.288)	(0.076)	(0.759)	(0.246)
UK			0.321 ***	0.453	0.687	0.488
			(0.105)	(0.202)	(0.343)	(0.357)
Slope Change x Country
Australia			1.251 *	0.982	1.298 *	0.990
			(0.137)	(0.158)	(0.143)	(0.160)
Korea			1.611 ***	0.860	1.818 ***	0.878
			(0.209)	(0.160)	(0.243)	(0.164)
USA			1.581 ***	1.156	1.650 ***	1.135
			(0.201)	(0.217)	(0.211)	(0.215)
Russia			1.553 ***	1.319	1.619 ***	1.342
			(0.172)	(0.236)	(0.181)	(0.242)
Switzerland			1.375 **	1.434	1.449 **	1.367
			(0.166)	(0.274)	(0.178)	(0.264)
UK			1.457 ***	1.034	1.528 ***	1.018
			(0.151)	(0.170)	(0.160)	(0.167)
Intercept Change x					4.532 ***	0.842
Children in HH					(2.020)	(0.620)
Intercept Change x Children in HH x Country
Australia					0.242 **	1.305
					(0.127)	(1.043)
Korea					1.783	1.965
					(1.291)	(1.796)
USA					0.338	1.664
					(0.199)	(1.518)
Russia					0.319 *	0.782
					(0.174)	(0.662)
Switzerland					0.892	0.730
					(0.542)	(0.723)
UK					0.486	1.206
					(0.240)	(0.981)
Covariates
Education (ref. higher)
Lower	1.421	1.432	1.379	1.008	1.265	0.987
	(0.272)	(0.279)	(0.349)	(0.360)	(0.321)	(0.356)
Medium	1.068	1.139	1.193	0.987	1.120	0.946
	(0.158)	(0.172)	(0.213)	(0.308)	(0.201)	(0.301)
Weekly hours worked	0.956 ***	0.950 ***	0.946 ***	0.950 ***	0.944 ***	0.949 ***
	(0.005)	(0.005)	(0.006)	(0.008)	(0.006)	(0.009)
Weekly hours worked (sq)	1.000 ***	1.000 ***	1.000 ***	1.000 ***	1.000 ***	1.000 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Socioeconomic incex (ISEI08)	0.987 ***	0.987 ***	0.991 **	0.980 ***	0.991 **	0.981 ***
	(0.003)	(0.003)	(0.003)	(0.004)	(0.003)	(0.004)
Age (ref. 18–25)
26–35	0.701 ***	0.734 **	0.754 *	0.763	0.770 *	0.745
	(0.068)	(0.073)	(0.100)	(0.122)	(0.102)	(0.120)
36–45	0.760	0.780	0.884	0.705	0.829	0.717
	(0.114)	(0.119)	(0.177)	(0.178)	(0.168)	(0.183)
46–55	0.947	0.925	0.875	1.100	0.788	1.109
	(0.193)	(0.191)	(0.233)	(0.373)	(0.212)	(0.381)
56–65	1.224	1.183	1.044	1.526	1.220	1.502
	(0.341)	(0.334)	(0.386)	(0.700)	(0.452)	(0.694)
Children in HH	2.019 ***	1.558 ***	1.367 **	1.843 ***
	(0.138)	(0.110)	(0.140)	(0.191)
Observations	16,456	16,456	10,422	6034	10,422	6034
Number of respondents	2493	2493	1595	898	1595	898
LR chisq	775 ***	1073 ***	1066 ***	220 ***	1130 ***	255 ***
DF	13	16	31	31	44	44
Log likelihood	−5719	−5570	−3363	−2099	−3332	−2082
AIC	11,464	11,172	6789	4261	6752	4252
BIC	11,564	11,295	7014	4469	7071	4547

Notes: * p < 0.05 ** p < 0.01 *** p < 0.001; Reference country for separate estimations and interactions of the “country”-variable is Germany. Children in the household are younger than 15 in the UK and younger than 17 in the other countries.

to the Appendix A and will show the results in this section in a more straightforward manner. In the hope of making the results table in the appendix slightly more interpretable, we have divided it into three parts. The first part shows the parameters related to poverty risks prior to the separation, the second part those associated with poverty risks afterwards, and the third part the covariates whose effects are assumed not to be altered by the separation. One exception here is the measure for children in the household. In the first three models this measure was entered as a covariate, but the final model was specifically designed to test this assumption. A final note of clarification for this table is that the first two models were estimated on the pooled sample, while the final two models were estimated separately for men and women.

Our first, two pronged hypothesis was that the effects of relationship dissolution on poverty risks are transitory, rather than persistent. Results of the first model in Table A1 in the Appendix A show that the odds of being at risk of poverty are three times (OR = 3.229; p < 0.001) higher immediately after separation than they would have been had the pre-separation trend continued. The interaction term of the time variables shows that, for each additional year, the odds of being at risk of poverty decrease by a factor of 0.9, confirming both parts of our first hypothesis.

In order to evaluate our second hypothesis, the second model in Table A1 introduces an interaction between the time variables and an indicator for gender. The results show that the baseline estimates are no longer significant, indicating that there is neither an immediate increase in the risk of poverty nor a change in the risk of poverty afterwards for men. This is not the case for women, for whom the odds of being at risk of poverty are five times (OR = 5.069; p < 0.001) higher immediately after separation. The trajectory effect shows a decline in the poverty risk with a yearly factor of 0.9 (p < 0.001), which does not significantly differ from the first model, although it is significantly different for women than for men. These results confirm our second hypothesis and justify separating the analyses by gender in order to evaluate the final two hypotheses.

The third hypothesis cannot readily be evaluated from Table A1. Instead, model 3 was estimated with a different reference category for country in order to obtain the ratios and confidence bounds for the slope and intercept effects of each individual country. Figure 2 and Figure 3 show the estimated intercept and trajectory effects, respectively. To improve the readers’ understanding of what these results signify, Figure 4 shows the average estimated trajectories of the AROP separated by country and gender. Apart from highlighting the differences between men and women, these figures also show considerable differences between countries. For women, the parameters indicating the transitory nature of the effect of relationship dissolution on being at risk of poverty (i.e., an intercept odds ratio higher than 1 and a slope odds ratio lower than 1) were only found to be significant in Australia and Germany. While the Swiss and UK women’s estimators are not significant, they are substantively closer to the estimates for Australia than the estimates for Korean and US women. For German men, we find that the odds of being at risk of poverty are 3.9 times higher immediately after separation. For men in the UK, this is 1.7 times higher. The only significantly negative association with poverty risks is found for men in the United States, where the odds of being at risk of poverty are two times lower immediately after separation. Looking at the trajectory effects for men, we see significantly positive effects in Switzerland and Russia. For both countries there was a significant decline in the odds of being at risk of poverty prior to the separation. These were, however, reversed into an overall increase in the risk of being at poverty after separation, by a factor of 1.12 per year.

Focussing on the immediate increases in poverty risks for women, three clusters can be distinguished. By cluster, we mean countries for whom the intercept effects differ significantly from other countries, but not from each other. We distinguish the relatively low (Korea, USA, Russia; OR < 5), moderate (UK, Australia; 5 < OR < 10) and relatively high (Germany, Switzerland; OR > 10) clusters. In the high cluster, the increase in the odds of being at risk of poverty are around three times that of the low cluster. Taken together, these results confirm our third hypothesis.

That these results need to be interpreted with necessary caution as to their overall meaning can be seen in Figure 4. The average estimated trajectories for women show the transitory effect of relationship dissolution on the risk of poverty more or less in all countries. Two possible exceptions here are Korea and the USA where, at the end of the observation period, it does not appear that the poverty risks have declined back to pre-separation levels, nor that they are even converging back to that level. Furthermore, these graphs illustrate that, while the immediate effect of separation on poverty risks is highest for German women, the transition back to pre-separation levels is also much more rapid than in, for instance, the UK, where the immediate effect is much lower. The graphs further illustrate that there are significant country differences in the effects for men between countries as well. The probability of being AROP for men can be said to be fairly stable in all countries, at least more stable than for women in most cases. In Germany, and to a lesser extent Australia and the UK, a transitory pattern, similar to that of women in most countries, was observed.

Finally, we evaluate our fourth and final hypothesis in the same way as the third hypothesis. Figure 5 shows the intercept effects of having children living in the household for each of the separate countries. Again, the effects appear to be gendered, with none of the countries showing significant effects for men. For women, the effects are significantly positive in Korea, Switzerland, and Germany. Considering the large confidence bounds for the estimates for Korean women, we cannot unambiguously confirm our fourth hypothesis.

5. Discussion and Conclusions

In the current study, we attempted to add to the literature on the consequences of divorce over the life course. Based on insights from life course studies (Dewilde 2003; Elder 1995) or the social stratification framework (Breen 2005; Shildrick and Rucell 2015; Townsend 1954), we follow more recent strategies to incorporate both the transitory and the durability approaches to poverty (Härkönen and Dronkers 2006; Oris et al. 2017; Vandecasteele 2015; Whelan and Maître 2008). By comparing data from harmonized surveys from seven long-running panels (Turek et al. 2021), we were able to analyse poverty risks following relationship dissolution against the backlight of different social policies.

We used fixed effects logistic models for estimating poverty risks in the 17 years surrounding divorce. These models control for time invariant unobserved characteristics associated with social stratification. By further controlling for time invariant effects of social stratification with several time-varying measures, we obtained a more precise estimate of the transitory effect of relationship dissolution. Our results are mostly in line with De Vaus et al. (2017), who analysed the economic consequences of relationship dissolution in six of our seven countries. However, while that report found no differences in pre- and post-separation equivalised household income for men in any of the countries, our relative measure of being at risk of poverty does show that UK and German men experience an increase in poverty risks immediately following separation, albeit much smaller than women. This result is more in line with the findings of McManus and DiPrete (2001) for the United States. However, looking at Russia, Switzerland, the United States and Korea, where men’s probability of being AROP decreased at the time of separation, there appears to be some indication that their poverty risks were actually increasing prior to separation, possibly pointing towards reversed causality where increased poverty risks increase the probability of relationship dissolution, rather than the other way around. Running the same regressions as were used to test hypothesis 3 but adding a lagged measure of poverty also points in this direction. The estimate was not found to be significant in the sample of women (1.04, p = 0.527) but it was for men (1.33, p = 0.003). While it is true that poverty acts as a stressor and as such increases the probability of divorce (Ambert 2005), the gendered finding in our models warrants further investigation, as it seems counterintuitive that poverty risks for women would not increase as they do for men

Although the evolution of poverty risks surrounding separation is quite similar across our sample, there are considerable differences between countries. We found that, for women, there were three categories of countries where the immediate effect after separation was either relatively high (Germany, Switzerland), moderate (United Kingdom, Australia), or low (Korea, United States, Russia). Looking at the comparison we made of different social policies on child and spousal support used by these countries, it is not clear why this would be the case. On the one hand, the United Kingdom and Australia have very similar regulations for child support, but while in the UK parents are able to get an advance on child support payments, which can be seen as a way to mitigate immediate poverty risks, this is not the case in Australia. Similarly, based on these social policies, there is no clear discernible reason why Korea, the United States, and Russia are similar in this respect.

Another notable difference between countries is that the transitory nature of increased poverty risks after separation for women was not found in the USA and Korea. While we are unable, from our models, to conclusively determine whether or not the immediate additional vulnerability for these women would persist over time, plots of the average poverty risk trajectories appear to point in this direction. Again, there is no apparent similarity between the two countries in terms of child and spousal support that could readily explain these findings.

Furthermore, although—contrary to Australia—both Switzerland and Germany do have a system of advanced payments in place, the immediate poverty risks increase the most in these countries. This seems counterintuitive. It could be that, rather than mitigating the immediate increase, advanced payments help in reducing the immediate effects more expeditiously, but again this is not shown in the analyses. Perhaps there are differences in the relative amount of child support payments between countries, or differences in compliance of expartners with the regulations that might explain the differences. Another possibility is that, while the seven countries have somewhat different outcomes relative to one another, they are still quite similar. Perhaps what sets them apart in these analyses would vanish in an even larger comparative framework including, for instance, Scandinavian, Latin American, or Central Asian countries.

Our results on the effect of children can similarly not readily be explained by differences in child support regulations between countries. Both Australia and the United States showed no immediate effects of having children in the household on poverty risks, while it increased this risk in the other countries. On the one hand, this might be due to the fact that there is considerable variation in these schemes between states in the US. On the other hand, a more fine grained measure of the number of children in the household would shed more light in this issue. Unfortunately, due to data constraints, we were not able to control for this.

Next, due to the coarseness of some of our measures, there are several other limitations to this study. First, as with all fixed effects estimations, while we are able to control for unobserved heterogeneity due to time invariant confounders, our models might still suffer from endogeneity due to time varying characteristics that are associated with both relationship dissolution and poverty risks that were not observed in this study. One important objection to trying to explain the observed differences through the schemes of spousal and child support that are in place in these countries is that, while these schemes might de jure seem very similar, this tells us nothing about their practical application and compliance with these regulations. For instance, Yiyoon Chung and Kim (2019) found that Korean mothers were considerably less likely than their American counterparts to receive child support from noncustodial partners. Future comparative research could then try to include measures of factual, rather than legal, differences in child support in order to try and explain the differences between welfare systems. Therefore, while our models allow us to state with some confidence the causal claim that relationship dissolution immediately increases the probability that women will be AROP in these countries, we cannot do the same for the mechanisms through which there are differences between countries, nor those between men and women. Furthermore, while, to the best of our knowledge, there have been no major changes in family law after 2011 (the earliest year of separation in our analyses) for most countries under observation, this is not the case for Switzerland and the UK, both of whom saw important changes in child custody and child support laws in 2014. Whether or not these changes play a significant role in the poverty risks for all those affected by relationship dissolution is a question for future research. Finally, on this limitation, it is important to note that there is a strong argument to be made that there is enough heterogeneity in family laws and regulation to warrant a between state analysis rather than being included as a single country. Unfortunately, this was not possible with the harmonized data from CPF that were used.

Additionally, as was pointed out in Section 2, it is an understatement to say that there is “considerable disagreement” in the literature on how poverty should be measured. While the AROP measure is useful to compare poverty risks within a country, it can be argued that it is not the best choice for comparisons across countries. Again, due to data limitations, we opted for this measure in order to get a bird’s eye view on emerging patterns, rather than a concise estimation of a concept as apparently elusive as “poverty”.

Furthermore, attrition bias and nonresponse is always a problem when studying relationship dissolution (De Vaus et al. 2017). In order to control for possible selective dropout related to poverty risks, we ran additional analyses and found no differences in the poverty risks prior to dissolution between those who were not observed in the first wave following the dissolution and those who were not. We also ran alternative analyses for those who had at least two observation points after the dissolution and found no significant differences with the estimates reported here. We ran similar robustness checks excluding 113 respondents for whom only one observation prior and one post separation was recorded. Again, the results did not differ significantly from those reported here.

Finally, while we found significant differences between countries in this study, it could be that, on a global scale, they are all members of the same broad class of countries. Since we found no obvious differences in the social policies for child and spousal support, there are two pathways for future comparative research into this topic. On the one hand, delving deeper into the subject by using more fine grained measures of social policies could uncover exactly what drives these differences. On the other hand, increasingly common longitudinal household data could allow researchers to expand the pool of countries to include (many) more countries that have, up until now, been underrepresented in these studies. By doing so, we would get a better answer to the question how different these countries really are.