Broken homes and empty pantries: the impact of couple separation on living standards in France

Mink, Julia Article — Published Version Broken homes and empty pantries: the impact of couple separation on living standards in France International Review of Economics Provided in Cooperation with: Springer Nature Suggested Citation: Mink, Julia (2025) : Broken homes and empty pantries: the impact of couple separation on living standards in France, International Review of Economics, ISSN 1863-4613, Springer, Berlin, Heidelberg, Vol. 72, Iss. 2, https://doi.org/10.1007/s12232-025-00499-6 This Version is available at: https://hdl.handle.net/10419/323902 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by/4.0/ Vol.:(0123456789) International Review of Economics (2025) 72:26 https://doi.org/10.1007/s12232-025-00499-6 RESEARCH ARTICLE Broken homes andempty pantries: theimpact ofcouple separation onliving standards inFrance JuliaMink1 Received: 3 September 2024 / Accepted: 30 June 2025 © The Author(s) 2025 Abstract This study sheds new light on the impact of couple separation on the living standards of women by considering the effects of separation on measures reflecting the adequacy of food consumption in addition to more commonly studied income and expenditure measures. Using panel data from France, I examine changes in disposable income, food expenditures and quantities purchased, diet quality and body weight around the time of separation and up to 8years later, compared to a control group of households that did not separate. Living standards decline to such an extent that some households cannot maintain their food consumption, resulting in weight loss. Keywords Separation· Divorce· Living standards· Income· Food consumption· Weight loss· Event study JEL Classifications I12· J12· D12 1 Introduction Family dissolution due to couple separation has become an event that a large part of the population will experience at some point in their lives. In France, the share of cohabiting couples who broke up their first union after less than 8years more than doubled, from 12% for unions formed in the 1970s to 29% for unions formed between 1997 and 2005 (INSEE 2015). A couple who married in 1970 has a 12% chance of ever getting divorced, compared to 44% for couples who married in 2014 (INSEE 2016). Trends are similar in many countries around the world (OECD 2011). Separation has many consequences for those involved, especially financially. Divorce has been associated with a reduction in income (see for example McKeever * Julia Mink [email protected] 1 Department ofEconomics, Institute forApplied Microeconomics, University ofBonn, Lennéstraße 43, 53113Bonn, Germany J.Mink 26 Page 2 of 56 and Wolfinger (2001); Avellar and Smock (2005); Tach and Eads (2015)). In France, it is estimated that women’s average standard of living drops by 14–25% in the year following a divorce, whereas men’s living standards are estimated to fall by 3–7% (Costemalle 2017). Since separation means the end of the economies of scale associated with cohabitation and the sharing of certain expenses, a decline in living standards after separation is to be expected. What remains an open question, however, is the extent to which the income shock of separation affects households’ ability to maintain a minimum level of consumption necessary to meet basic needs. The studies that look at the effect of separation on income provide information on relative changes in the financial situation, but not so much on whether households have difficulty meeting their basic needs in absolute terms. Households could vary in their ability to adapt to lower budgets by being more or less able to turn to less expensive options. The question of households’ ability to maintain a minimum level of consumption is important. When households are unable to meet their basic needs, there can be costly negative impacts on human capital, such as poorer health. Lower economic resources have been associated with poorer adult and child outcomes, including poorer mental and physical health and lower educational attainment (McLanahan etal. 2013; Tach and Eads 2015). This study addresses this gap in the literature by examining whether couple separation affects households’ ability to meet basic consumption needs, with a focus on food consumption as a fundamental necessity. More precisely, this study estimates the effects of couple separation on household disposable income, food expenditures and quantities purchased, diet quality, and body weight. Looking at the quantity of food purchased, diet quality, and body weight is crucial because it is more informative about whether households are having difficulty meeting their basic dietary needs than changes in income and even food expenditures would suggest. A fall in income and food expenditures does not necessarily signal a decrease in actual food consumption as households might be able to shop for bargains and substitute for cheaper products instead. A major obstacle to studying this issue is the difficulty in obtaining individuallevel data on household composition and income, as well as reliable and detailed data on food consumption. For this study, I use consumer panel data from Kantar Worldpanel for France, which include basic demographic information on household members, household composition, and detailed information on food purchases entered by households via scanners after each grocery purchase. Although men are included in the sample, most post-separation observations involve female-headed households. The analysis therefore concentrates on women, which is also particularly relevant given existing evidence that they tend to bear the financial consequences of separation. In an event study approach, I examine how the outcome variables evolve in the years before and after the separation, relative to the period just before separation and compared to households where no separation is observed. I control for time-invariant differences across households and the impact of macrolevel shocks by including household and year fixed effects. To identify vulnerable households and explore potential adaptation mechanisms, the analysis also examines heterogeneity by pre-separation income, family composition, and employment Broken homes andempty pantries: theimpact ofcouple separation… Page 3 of 56 26 status. I perform several robustness checks by selecting control households differently, omitting control households altogether and calculating estimates that are robust to treatment effect heterogeneity, as proposed in the recent two-way fixed effects literature. Several key findings emerge from this study. First, I find that women suffer significant and permanent losses of economic resources after separation. In the first year after separation, their disposable income is around 20% lower relative to pre-separation levels and remains 30% lower until the end of the observation window up to 7years after the separation. Second, women appear to cut back on food consumption after a separation. Food expenditures and quantities purchased decline by 20–40% relative to pre-separation levels. Third, the changes in income and food consumption are accompanied by a 1.5% decline in the women’s body mass index (BMI1) during the first 3years after separation. I find evidence for an increase in the probability of being underweight (BMI of less than 18.5) as well as being of normal weight, while the probability of being overweight (BMI of over 25) decreases. The weight loss occurs despite an increase in the share of unhealthy food products purchased consisting of more salty, sweet, fatty, and convenience foods, which are calorie-dense. Importantly, the effects on income and food purchases are not simply a mechanical consequence of the departure of a household member. If, instead of adjusting for household composition, per capita income and food purchases are used, the results are qualitatively equivalent. I find no effect when an adult other than the partner (for example, an adult child) leaves the household. If the effect were purely mechanical, I would expect to find similar results when any adult household member leaves the household. Fourth, the effects on income are stronger in high-income households, but food purchases and body weight decline more strongly in low-income households and single-parent households. The probability of being underweight after separation increases in low-income households, but not in high-income households. Labour market participation does not appear to significantly buffer the effects of separation. In contrast, entering a new relationship seems to mitigate the impact, although the estimates are imprecise. I also find effects for men, though they are more moderate. Income declines by approximately 8%, and food purchases fall by 10–20%, though these estimates are not statistically precise. This study contributes to the literature on the economic consequences of separation in three key ways. First, it goes beyond commonly used indicators such as income and expenditures by incorporating direct measures of nutritional well-being, including food quantities purchased, diet quality, and body weight. These outcomes provide a more complete picture of living standards and vulnerability, recognizing that households may adapt to lower budgets without immediately reducing calorie intake or nutritional adequacy. Second, it provides medium and long-term estimates of separation effects using detailed longitudinal data on household consumption. While many studies have documented short-term income declines following divorce, 1 Defined as body weight divided by height squared. A commonly used measure to standardise weight for height and a rule of thumb to broadly categorise a person as underweight, normal weight, overweight, or obese. The results are qualitatively the same when using body weight directly. J.Mink 26 Page 4 of 56 typically ranging from 23% to 40% in the year after separation (Hoffman 1977; Duncan and Hoffman 1985b; Bianchi and McArthur 1991; Holden and Smock 1991; McLanahan and Sandefur 1994; Peterson 1996; Galarneau and Sturrock 1997; McKeever and Wolfinger 2001; Avellar and Smock 2005; Tach and Eads 2015), fewer studies examine longer-term consequences. Where such studies exist, they often rely on dated samples or limited covariate adjustment (Weiss 1984; Duncan and Hoffman 1985a; Fisher and Low 2009; De Vaus etal. 2014). This study addresses these gaps by tracking outcomes over several years and accounting for both time-varying household characteristics and unobserved heterogeneity. Third, it provides new evidence from France, a context that differs significantly from the United States, where most prior research has been conducted. In 2017, public spending on family benefits in France, including cash transfers, services, and tax relief, represented over 3.5% of GDP, compared to only 1% in the United States (OECD 2017). This difference makes France a valuable setting for examining whether more extensive social protection alters the consequences of separation. This study is most closely related to Page and Stevens (2004), who examine the effects of family structure on income and food spending in the United States, and to Bonnet etal. (2021), who study divorce effects on living standards and labour supply in France. The present analysis builds on this work by incorporating nutritional and anthropometric outcomes, offering a broader perspective on how separation affects household well-being. Overall, the study provides evidence that couple separation leads to a significant and lasting decline in living standards for women, particularly those in lowincome and single-parent households. Reductions in food quantities purchased and body weight suggest that some households struggle to meet basic nutritional needs, even in a country with a comparatively generous social safety net. While there is also some evidence of adverse effects for men, these are more moderate and less precisely estimated. Taken together, the findings highlight the importance of going beyond income-based indicators to identify vulnerability and suggest that policy support may be needed to ensure that all households, especially the most economically fragile, can maintain access to adequate nutrition and basic living standards following separation. The remainder of the paper is as follows. Section 2 describes the data and details the empirical strategy, Sect.3 presents and discusses the results, and Sect.4 concludes. 2 Method 2.1 Data This study draws on detailed data from a representative sample of French households collected by Kantar Worldpanel, covering the period from 2005 to 2014. The dataset is well suited to this analysis because it combines rich information on household characteristics and income with detailed records of food purchases, allowing for the construction of both financial and consumption-based measures of living standards. Kantar Worldpanel is a private firm specializing in consumer panels and Broken homes andempty pantries: theimpact ofcouple separation… Page 5 of 56 26 market research, similar to Nielsen in the United States. Participating households are equipped with scanners to record food purchases with barcodes. Items without barcodes are entered manually by household members.2 The dataset provides information on individual and household characteristics, including household composition, household disposable income bins where income includes any income and transfers (earnings, pensions, alimony payments, government transfers), the socioprofessional category, age, sex, height, weight, education level, and labour market status of each household member. Information on food purchases includes product type, quantity, price and purchase date. As the data on individual and household characteristics are updated annually, the time interval for analysis in this study is the year. I construct household food quantities purchased as the sum of all products purchased annually, measured in kilograms, and household food expenditure as the sum of annual food expenditures, expressed in euros. The share of unhealthy food products is defined as the ratio of the annual quantity in kilograms of prepared food products (such as pizza, sauerkraut, cassoulet), salty-fatty products (such as finger food, crisps, crackers, appetisers), and sweet-fatty products (such as candy, chocolate, cookies, pastries, ice creams, jams) to the total amount of annual food purchases, also measured in kilograms. I use information on body weight and height to calculate the body mass index (BMI) of each household member. BMI is defined as weight divided by height squared. It is a commonly used measure to standardise weight for height and serves as a rule of thumb to categorise individuals as underweight (BMI under 18.5kg/ m 2 ), normal weight (18.5–24.9), overweight (25–29.9), or obese (30 or more). BMIs under 20 and over 25 have been associated with higher all-cause mortality, with the risk increasing with distance from the 20–25 range. I do not have data on food consumed outside the home, but households report the number of meals typically eaten at home for each day of the week. I use this information to calculate the weekly average number of meals eaten at home in a given year, in order to investigate whether separation affects eating habits at home as a potential mechanism behind the observed changes in food purchases. A limitation of the Kantar Worldpanel data is that it does not include direct information on the marital or partnership status of household members. Households are classified into young, mid-age and old single or couple households and family households. The status changes from couple to single household when a partner leaves, but family households remain classified as family as long as children are present. To overcome this limitation, I exploit the fact that each individual is assigned a status code that is not reassigned to other household members in case the member leaves the household. Status 1 is always assigned to the female household member who is in charge of the household, as opposed to status 3, which designates a dependent female, such as a child. Status 2 is always assigned to a male household member who is assumed to be in charge of the household, as opposed to status 4, which denotes a dependent male, such as a child. I define separation as the departure 2 For more information, see the Kantar Worldpanel website https:// www. kanta rworl dpanel. com/ global/ Consu merPanels. J.Mink 26 Page 6 of 56 from the household of an individual with status 1 or 2. I am therefore not distinguishing between the separation of cohabiting or married couples. This approach may also capture some instances of partner death rather than separation. However, this is likely to be rare in the working-age population. In France, only about 6% of single-parent families result from the death of a partner, and fewer than 5% of family dissolutions before age 45 are due to death rather than separation (INSEE 2015; INED 2018). I show in Sect.3.2 that results are robust when restricting the sample to individuals younger than 45years, for whom partner death is an unlikely source of bias. I observe a total of 42,000 households for an average duration of 5.9years, while the maximum observation window is 10years. The results presented in this study are based on analyses using a sample of treatment households that is unrestricted, except for the requirement that the household be observed at least one year before separation, in the year of separation, and at least one year after separation. This ensures that the effects of separation can be identified. As this restriction has no effect on the estimated coefficients, I avoid imposing further limitations on the sample of separating households, in order to retain as many separation events as possible. The control group is restricted to households observed for at least 7 consecutive years. This restriction aims to avoid the appearance of pre-trends in food purchases. Compared to other outcome variables, food purchases vary more strongly across households and over time. Using a panel composed of households observed for short periods may introduce sample composition effects that could bias the results. Nonetheless, the results are qualitatively similar even without this restriction on the control sample. See Sect.3.2 for further discussion and additional sensitivity analyses using alternative event time windows to verify that results are not driven by changes in sample composition. Table1 reports summary statistics for both the unrestricted sample and the subsample restricted to households observed for at least 7 consecutive years, with observations pooled across households and years. The two samples are broadly similar, although the restricted sample has higher average food purchases and older household members. Households included in the panel for only short periods may report food purchases less consistently. The higher average age in the restricted sample may reflect different time constraints, with older individuals potentially having more availability to remain in the panel. The final sample comprises 10,033 households, observed for an average of 8.6years. Table 2 provides national statistics for France to compare with the study data. In 2010, average household disposable income in France was EUR 2919, which is close to the sample average of EUR 2646. The inactivity rate in the sample is also similar to the national level. Household size in the sample is slightly larger (2.6 vs. 2.3 in the general population). The average age, particularly of men, is somewhat higher in the sample. However, the results remain robust when restricting the sample to individuals under age 45 (see Sect.3.2 for results and discussion). Table3 compares summary statistics for treatment households in the year before separation with non-single households in the control group. In separating households, both income and food quantities purchased are lower, and household members are younger on average than in the control group. For this reason, I apply household Broken homes andempty pantries: theimpact ofcouple separation… Page 7 of 56 26 fixed effects and adjust for partner age in all regressions. The results are also qualitatively similar when the control group is omitted entirely or constructed differently, including using propensity score matching (see Sect.3.2). Out of the 1038 households in which a separation is observed, 854 involve the departure of a male partner (individual with status 2), and 184 involve the departure of a female partner (status 1). The “remaining” household is defined as the one in which the partner continues to respond to the Kantar Worldpanel survey. In Kantar Worldpanel, the panelist responsible for grocery shopping and reporting household purchases is typically a woman, and is often the one who continues to be observed after separation. At the time of separation, 475 of the households are couple-only households. The remaining 563 include at least one additional household member. Among them, 306 households include at least one child under age 18, and 206 include at least one child under age 12. The rest are composed of other adult members in addition to the separating couple. Table 4 shows the number of households observed at each year relative to the year of separation. Since not all households are observed throughout the full 10-year window, the sample composition varies by event time. A total of 1038 households are observed in the year before, during, and after separation, but fewer observations are available in earlier and later years. The results remain qualitatively similar when households with fewer observations are gradually excluded, and when applying the estimator proposed by De Chaisemartin and d’Haultfoeuille (2020), which accounts for unbalanced panels. See Sects.3.2 and 3.3 for further results and discussion. 2.2 Empirical strategy To estimate the effects of couple separation on household living standards, I use an event study framework that tracks the evolution of outcome variables before and after the separation. The analysis compares households that experience a separation to those that do not, and contrasts outcomes each year to the year immediately preceding the separation. Specifically, I estimate the following model: where Yht denotes the outcome for householdh in yeart, 𝛼h and 𝛾t are household and year fixed effects, Dj ht denotes the treatment indicator for a separation occurring j∈[j,  j ] periods away from t,and Xht is a vector of time-varying household characteristics. The error term 𝜖ht is clustered at the household level. The variables of interest are the event time coefficients 𝛽j , which measure the impact of separation relative to the households that do not separate, and, since the event time dummy j=−1 is omitted, relative to the year just before the separation. The outcome variables include household disposable income, food expenditures, the quantity of food purchased in kilograms, the share of unhealthy foods in total purchases, and the body weight of household members. While income and expenditure are commonly used measures of living standards, they may not fully Y ht =𝛼h+𝛾t+  j ∑ j=j 𝛽jDj ht +𝜌Xht +𝜖ht , J.Mink 26 Page 8 of 56 reflect a household’s ability to meet basic needs. Income can underestimate actual resources, and a reduction in expenditure may not imply reduced consumption if households switch to cheaper alternatives. Measures of food quantity, diet quality, and body weight offer a more direct view of nutritional sufficiency and material hardship. Following Page and Stevens (2004), the main specifications use the log of income and food purchases, while controlling for household size in Xht . This accounts for changes in household composition from the arrival or departure of members in both treated and control households. The estimated event time effects therefore capture the additional impact of losing a partner, over and above the general effect of a change in household size. In robustness checks, I estimate models using per capita outcomes and consumption unit adjustments. In the latter, the firstadult counts as 1, additional individuals aged 15 or older as 0.7, and children under 15 as 0.5, to reflect age-based needs and economies of scale. The results are robust to these alternative specifications. Identifying causal effects of couple separation is challenging because separation is unlikely to be exogenous. First, households that experience a separation may differ systematically from those that do not. Second, time-varying shocks may both increase the likelihood of separation and directly influence the outcome variables. To address potential biases, I include both household and year fixed effects. Household fixed effects absorb all time-invariant differences across households, and year fixed effects account for macroeconomic shocks common to all households. The regressions also include households that do not separate as controls, a standard feature in event study designs that improves statistical precision by leveraging both the timing of events and comparisons with untreated units (Miller 2023). While the event time coefficients are identified using within-household variation in the treated group, year fixed effects and the coefficients on time-varying controls are identified using variation from both treated and control households. To assess whether differences between treated and control households might bias results despite fixed effects, I conduct two further robustness checks. First, I reweight the control group using propensity scores to make it more comparable to the treated group on observable characteristics. Second, I estimate models using only treated households, relying exclusively on the timing of separation. Both approaches produce results that are qualitatively consistent with the main findings. To address concerns about time-varying unobserved shocks, I test for pre-trends in the outcomes. If, for example, a health shock reduces income and later leads to separation, I would expect to see a decline in income or food purchases before separation. In the data, I find no such pre-trends in income, food quantities, or body weight. There is a modest increase in the share of unhealthy food purchases before separation, but other outcomes remain stable. This supports the view that the estimated effects are not primarily driven by shocks occurring before separation. Nevertheless, I cannot rule out bias from shocks that simultaneously cause separation and affect outcomes within the same year. I return to this point in the discussion of results. In all regressions, I use the longest available event window to capture both shortand long-run effects. Treated households are observed for up to 7years before and Broken homes andempty pantries: theimpact ofcouple separation… Page 15 of 56 26 % difference relative to j -1 Years since break-up 30% poorest households % difference relative to j -1 Years since break-up 30% richest households A. Income % difference relative to j -1 Years since break-up 30% poorest households % difference relative to j -1 Years since break-up 30% richest households B. Food quantities purchased % difference relative to j -1 Years since break-up 30% poorest households % difference relative to j -1 Years since break-up 30% richest households C. Spouse's body weight index Difference relative to t -1 Years since break-up 30% poorest households Difference relative to t -1 Years since break-up 30% richest households D. Share of unhealthy food purchased Fig. 2 Trend in outcome variables around separation, by pre-separation per capita income. The figure shows event time coefficients for the treatment group of households where the male partner leaves the female partner, relative to the control group of households where no separation occurs and relative to the year just before separation ( j =− 1 ). The lefthand panels present results for the 30% of households with the lowest pre-separation per capita incomes, while the right-hand panels show results for the 30% of households with the highest incomes. The dependent variables are the logarithm of income, food quantities purchased, remaining partner’s BMI, and the share of unhealthy food products purchased by household i in year t. In addition to household and year fixed effects and household size, all regressions include dummies for both partners’ ages. The data cover the period from 2005 to 2015. The 95% confidence intervals are based on standard errors clustered at the level of the household J.Mink 26 Page 16 of 56 also become statistically significant when using a single post-separation dummy to improve precision.4 The finding that income declines are more pronounced in households with higher pre-separation income is consistent with previous research. For example, Fisher and Low (2016) show that women from the highest-income households experience the largest and most persistent declines in their standard of living following separation. These high-income households may therefore appear more vulnerable. However, the sharper drop in income among high-income households likely reflects the higher earnings of the partner who leaves, rather than greater vulnerability. By contrast, the more immediate and severe reductions in food purchases and body weight among low-income households suggest that these households are less able to smooth consumption and may in fact bemore exposed to the economic shock of separation. Supporting this interpretation, I find that the probability of being underweight increases only among low-income households. In higher-income households, the shift in BMI distribution is driven by a reduction in the probability of being overweight or obese, rather than an increase in underweight status. See Fig.15 in the Appendix. This suggests that weight loss among low-income households may be less voluntary and more likely to reflect material hardship. This pattern is consistent with differences in saving capacity. Low-income households may have fewer financial buffers and less room to reallocate spending. In contrast, high-income households may have accumulated assets that help cushion the effects of separation. These results underscore the importance of looking beyond income to consumption-based measures when evaluating changes in well-being and identifying the most vulnerable populations. 3.3.2 Heterogeneity oftreatment effects inrelation tothepresence ofchildren Figure3 shows results separately for couple-only households (left-hand panels) and households with children under age 18 at the time of separation (right-hand panels). Income declines by around 20% in couple-only households, compared to a 40% decline in households with children. Food purchases also fall more sharply in family households, with reductions of up to 40%, whereas declines are more progressive in couple-only households. Women’s body weight decreases significantly in both groups, but the effect is almost twice as large in households with children. Overall, the effects of separation on women are more pronounced when children are present, although the difference is statistically significant only for income.5 Households with children are more likely to fall within the bottom 30% of the pre-separation income distribution than couple-only households. When examining treatment effect heterogeneity by family type within income groups, I find that the 4 Statistical significance is assessed using an interaction model that includes either event time dummies or a single post-separation dummy, interacted with a dummy for households in the top 30% of the preseparation per capita income distribution. A difference is considered statistically significant if the interaction term is significant at the 5% level. 5 Statistical significance is assessed using an interaction model that includes either event time dummies or a single post-separation dummy, interacted with a dummy for family composition groups. A difference is considered statistically significant if the interaction term is significant at the 5% level. Broken homes andempty pantries: theimpact ofcouple separation… Page 17 of 56 26 Percentage difference relative to t -1 Years since break-up Couple only households Percentage difference relative to t -1 Years since break-up Households with children A. Income Percentage difference relative to t -1 Years since break-up Couple only households Percentage difference relative to t -1 Years since break-up Households with children B. Food quantities purchased Percentage difference relative to t -1 Years since break-up Couple only households Percentage difference relative to t -1 Years since break-up Households with children C. Spouse's body weight index Difference relative to t -1 Years since break-up Couple only households Difference relative to t -1 Years since break-up Households with children D. Share of unhealthy food purchased Fig. 3 Trend in outcome variables around separation, by family composition (presence of children). The figure shows event time coefficients for the treatment group of households where the male partner leaves the female partner, relative to the control group of households where no separation occurs and relative to the year just before separation ( j =− 1 ). The left-hand panels present results for couple-only households, while the righthand panels show results for households with children. The dependent variables are the logarithm of income, food quantities purchased, remaining partner’s BMI, and the share of unhealthy food products purchased by household i in year t. In addition to household and year fixed effects and household size, all regressions include dummies for both partners’ ages. The data cover the period from 2005 to 2015. The 95% confidence intervals are based on standard errors clustered at the level of the household J.Mink 26 Page 18 of 56 impacts of separation remain generally larger in households with children than in couple-only households. However, the estimates are imprecise, and the differences are not statistically significant. See Fig.16 in the Appendix. Some of the observed differences in food purchases could reflect custodial arrangements. Children may be registered in the household even if they do not reside there full-time. In France, following separation, 76% of children are primarily cared for by their mother, 9% by their father, and 15% live in alternating residence arrangements. Even when not the primary caregiver, the non-custodial parent usually has regular visitation, often including weekends (Bonnet etal. 2015). The average number of meals eaten at home increases slightly after separation (see column 1 in Table9 and the section on mechanisms below), but this may be an imperfect proxy for children’s presence and consumption. While this may introduce some measurement error, the stronger effect on women’s body weight in households with children suggests a genuine reduction in overall calorie intake, consistent with a decline in food purchases that exceeds changes in household needs. Taken together, these results point to heightened vulnerability among single-parent households following separation. 3.3.3 Heterogeneity oftreatment effects bypre‑separation employment status Figure17 in the Appendix shows the effects of separation by the woman’s pre-separation employment status. The left-hand panels show results for women who were employed before separation, and the right-hand panels for those who were not. The decline in income is greater among employed women, while the drop in food purchases is more pronounced among women who were not employed before separation. This pattern is consistent with dual-earner households experiencing a sharper loss in income due to the departure of a high-earning partner, assuming similar partner earnings, while inactive women may face greater difficulty adjusting, as reflected in deeper cuts to food consumption. There is little difference in BMI loss across the two groups. The share of unhealthy food purchases declines among households where the woman was employed, while it remains largely unchanged in households where she was inactive. These differences should be interpreted with caution, as none are statistically significant.6 Labour market participation is also likely to change in response to separation and is discussed below as a potential mechanism for mitigating declines in living standards. 3.3.4 Heterogeneity intreatment effects due tounobserved factors Treatment effects may vary not only along observable characteristics but also in unobserved ways. If this is the case, the main estimates reflect a weighted average 6 Statistical significance is assessed using an interaction model that includes either event time dummies or a single post-separation dummy, interacted with a dummy for pre-separation employment status. A difference is considered statistically significant if the interaction term is significant at the 5% level. Broken homes andempty pantries: theimpact ofcouple separation… Page 19 of 56 26 of the underlying effects, where the weights may be unintuitive or undesirable. In particular, some weights may be negative, potentially biasing the average treatment effect (Sun and Abraham 2021; De Chaisemartin and d’Haultfoeuille 2020). To address this concern, I apply the approach proposed by De Chaisemartin and d’Haultfoeuille (2020), which uses not-yet-treated units and a parallel trends assumption to estimate treatment effects by group before averaging them. This method yields results that are qualitatively identical to those of the main analysis, as shown in Fig.18 in the Appendix. 3.3.5 Mechanisms: eating outversuseating athome, labour market participation, andrepartnering Several mechanisms may help explain the observed changes in living standards following separation. One concern is that the decline in food purchases might reflect a shift from eating at home to eating out, rather than a true drop in food consumption. Although food purchased outside the home is not recorded in the data, I observe the average number of meals eaten at home per week, which serves as a proxy for meals consumed at home versus outside the home. As shown in column 1 of Table9, this number increases after separation. This suggests that the decline in food purchases is not driven by substitution toward food outside the home. If anything, eating out appears to decrease, possibly due to its higher cost and lower appeal when dining alone. The concurrent drop in BMI further supports the interpretation that total food intake declines after separation. Labour market participation may also serve as an adjustment mechanism. As shown in column 2 of Table 9, the probability that the woman remains inactive falls after separation. Above, I showed that being employed prior to separation does not protect women from declines in income or food consumption. I now examine whether entering the labour market after separation helps mitigate these effects. Figure19 shows outcome trajectories for households in which the woman enters the labour market following separation. While estimates are less precise due to smaller sample sizes, the direction of effects suggests that labour market entry does not fully offset the economic consequences of separation. Household income declines by approximately 40–50%, and BMI decreases by 2.5% up to 5years after separation. Food purchases appear to decline by around 25%, although these coefficients are imprecisely estimated. These results should be interpreted with caution, as they likely reflect selection. Women who become economically active may be those most negatively affected, which suggests that remaining inactive would have led to even poorer outcomes. As such, the estimates should be viewed as descriptive evidence rather than causal effects. Repartnering is another potential adjustment mechanism. Column 3 of Table9 shows that the probability of entering a new relationship increases in years 2 and 3 after separation. To assess whether repartnering helps mitigate the negative effects of separation, I estimate treatment effects separately for households where a new partner joins. Figure20 shows that in households where a new couple forms, the negative effects of separation are less persistent. Income still declines by about 20% in the first 2 years after separation but returns to pre-separation levels thereafter. J.Mink 26 Page 20 of 56 Food purchases also appear to fall in repartnered households, but estimates are not statistically significantly different from 0 after year 3. While entering the labour market does not appear to mitigate the decline in living standards, forming a new partnership is associated with a quicker recovery in income. This finding is consistent with previous research, such as Page and Stevens (2004), which shows that remarriage reduces the long-term economic costs of separation. However, as with labour market participation, repartnering is likely subject to selection, and these results should be interpreted as descriptive evidence only. 3.4 Discussion This study finds that separation leads to persistent declines in the living standards of women in France. Disposable income falls by around 30%, food purchases drop by 20% to 40%, and body weight declines, consistent with reduced caloric intake. These changes persist for several years and are not driven by compositional effects. Reductions are particularly pronounced in low-income and single-parent households, which also show a higher probability of underweight status post-separation. Labour market participation offers limited protection. Women who were employed before separation experience smaller reductions in food purchases but still face losses. There is suggestive evidence that repartnering may help mitigate some effects, but estimates are imprecise due to the small number of observed cases. The evidence suggests that both partner income loss and loss of economies of scale play a role, and that neither transfers nor own earnings are sufficient to prevent the decline in living standards post-separation. These findings are broadly consistent with existing literature. In the United States, Page and Stevens (2004) document a 50% drop in disposable income and a 35% decline in food spending after separation. A reason the estimated income effects may appear more modest in this study is measurement error. Income is reported in bins, which likely introduces attenuation bias and pushes estimates toward zero. As a result, the estimated 30% drop in income may be a lower bound. This also cautions against using my results to infer an income elasticity of food demand. A 40% drop in food purchases from a 30% income fall would imply an elasticity greater than one, which is inconsistent with the empirical literature, where elasticities typically fall between zero and one. Another possible reason is that the effects are overall smaller in France due to more generous public transfers. Bonnet etal. (2021) report a disposable income decline of 18.5% for women, while Costemalle (2017) finds declines of 14–25%, depending on marital status. The effects of separation in France appear to last longer than in the US. I find little evidence of recovery over time, whereas Page and Stevens (2004) report that food purchases partially recover within 6 years. Page and Stevens (2004) attribute this recovery mainly to remarriage. I rarely observe a new partner joining the household, which may explain the lack of aggregate recovery in my sample. For households that do repartner, I also find that income reverts to pre-separation levels 4years after separation. Broken homes andempty pantries: theimpact ofcouple separation… Page 21 of 56 26 The decline in body weight is consistent with findings from some previous studies (Lee etal. 2004; Eng etal. 2005), though others report weight gain post-separation (Mata etal. 2018). I find no comparable estimates for changes in the nutritional quality of purchases. A few studies have explored post-separation dietary changes, though typically focusing on specific items (Lee etal. 2004; Vinther etal. 2016). Although the main results reflect outcomes for women and children, since most post-separation households in the data are headed by women, I also find modest effects for men. Income falls by around 8% and food purchases decline by around 18%. This is consistent with a broad literature showing more severe and consistent post-separation impacts for women, as in (Hoffman 1977; Duncan and Hoffman 1985b; Bianchi and McArthur 1991; Holden and Smock 1991; McLanahan and Sandefur 1994; Peterson 1996; Galarneau and Sturrock 1997; McKeever and Wolfinger 2001; Avellar and Smock 2005; Tach and Eads 2015), with more heterogeneous and generally weaker effects for men (Smock 1994; Galarneau and Sturrock 1997; McManus and DiPrete 2001; Bonnet etal. 2021). A key question is whether the results can be interpreted as causal effects of separation. The study design supports this interpretation in several ways. Household fixed effects account for all time-invariant differences, and time fixed effects control for common macroeconomic shocks. Robustness checks show that the findings are not driven by control group selection, time windows, or sample composition. Placebo tests using randomly assigned separation dates reveal no similar patterns, suggesting the observed effects are unlikely to be due to chance. I also find no similar declines when another household member, rather than the partner, leaves the household. Some bias from unobserved, time-varying shocks cannot be fully ruled out. For example, sudden health events could affect income, diet, and relationship stability simultaneously. However, the absence of pre-trends, the sharp timing and persistence of the changes point to separation as the driver of the results. Another central question is whether the observed effects are relevant from a policy perspective. One interpretation is that the decline in living standards following separation is so severe that many households cannot maintain minimum consumption levels to meet their nutritional needs. This is reflected in sharp reductions in food purchases and measurable weight loss. These findings may seem surprising. One might expect that in a country with a well-developed welfare system like France, public and private transfers would at least prevent households from falling below subsistence thresholds. If that is not the case, policy intervention may be justified both on ethical grounds and to prevent broader societal costs arising from unmet basic needs. An alternative explanation is that some women may intentionally lose weight after separation, for example to improve their chances of finding a new partner. This could help explain the decline in the probability of being overweight and the rise in the probability of being of normal weight. However, it does not account for the increase in underweight status, except in rare cases such as eating disorders. In addition, the strongest reductions in food purchases and body weight are observed among low-income and single-parent households, which are arguably less able to smooth consumption after an income shock. In these groups, the increase in underweight prevalence is not matched by similar patterns among higher-income households, J.Mink 26 Page 22 of 56 where weight changes appear to reflect a shift from overweight to normal weight. These patterns suggest that, at least for some low-income households, the reductions in consumption and body weight are driven by financial constraints rather than preference. Financial pressure may also influence behaviour indirectly. For women in economically insecure households, weight loss may be seen less as a personal goal and more as a strategy to improve prospects on the dating market, especially if repartnering is viewed as the only realistic way to restore living standards. Prior studies have found that households often regain their pre-separation standard of living only after repartnering (Page and Stevens 2004), and I show suggestive evidence of the same. Women from higher-income households may face less urgency to re-enter the dating market, given greater financial security. These dynamics suggest that some behavioural changes after separation may reflect economic necessity rather than personal preference. Mental health could also play a role. Separation may lead to depression or related challenges that reduce appetite and food intake. If mental health effects are more severe in financially strained households, this would reinforce the interpretation that economic vulnerability is central to the observed outcomes. Taken together, these findings indicate that the stronger effects observed in lowincome households are more likely to reflect reduced living standards than voluntary lifestyle changes. The fact that body weight loss reaches unhealthy levels in these households, alongside a rise in underweight prevalence, points to unmet basic needs. If the decision to seek a new partner is shaped by financial constraints, this raises concerns about constrained autonomy and economic dependence after separation. Some individuals may feel unable to leave unsatisfactory relationships or may enter new ones primarily for economic survival. Similarly, if mental health effects like depression are contributing to lower food intake, and are themselves exacerbated by economic strain, this provides an additional reason for policy intervention. The findings raise concerns about the capacity of welfare systems in high-income countries to protect households from falling below basic living standards after separation. They also show the importance of looking beyond income when assessing the effects of separation. Changes in food consumption, body weight, and diet quality offer more direct indicators of households’ ability to meet essential needs. The results suggest that some separated households, especially those with low incomes or headed by single parents, struggle to maintain adequate nutrition, with potentially serious implications for health and social policy. Policy intervention may be justified not only on ethical grounds but also for economic and public health reasons. The observed rise in underweight prevalence and the increase in the share of unhealthy food products purchased point to a deterioration in diet quality, which may lead to poorer health and lower productivity over the long term. While some individuals shift from overweight into the normal weight range, this cannot be seen as a health gain when it reflects involuntary weight loss Broken homes andempty pantries: theimpact ofcouple separation… Page 23 of 56 26 and worsening nutrition. It would be misguided to view reduced purchasing power as beneficial simply because it lowers calorie intake. A policy priority could be to increase support for single-parent households. Children, though not responsible for their family situation, are often part of the most severely affected households. Without adequate support, they risk lasting disadvantages in health, education, and well-being, with long-term costs for both individuals and society. More broadly, strengthening the social safety net would help ensure that all households can meet basic needs after separation. 4 Conclusion This study provides new evidence on the long-term effects of couple separation on women’s living standards in France, with a particular focus on food consumption and nutritional well-being. Using detailed panel data and an event-study design, I show that separation leads to sharp and persistent declines in disposable income, food purchases, and women’s body weight. These effects are especially pronounced in low-income and single-parent households, where the risk of underweight status increases. The findings challenge the view that welfare systems in high-income countries fully shield households from material hardship. In France, despite a generous safety net, some households struggle to maintain adequate nutrition after separation. Labour market participation offers limited protection, and while repartnering is associated with partial recovery, it may also reflect constrained choices shaped by economic need. More broadly, the results highlight the importance of looking beyond income when assessing the impact of separation. Using indicators such as food quantities purchased, diet quality, and body weight provides evidence that some households fall below subsistence thresholds following separation, with potential long-term consequences for health and well-being. Ensuring that households can meet their basic needs after separation is a matter of public health, social justice, and economic efficiency. Strengthening support for low-income, single-parent families and improving enforcement of existing support mechanisms could help prevent nutritional hardship and promote more resilient household trajectories following family dissolution. Appendix See Tables1, 2, 3, 4, 5, 6, 7, 8, 9 and Figs.4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19 and 20 J.Mink 26 Page 24 of 56 Table 1 Summary statistics, pooled household-year observations Summary statistics of pooled household-year observations using all available household observations, compared to the sample restricted to households observed for at least 7 consecutive years. The last column shows mean differences with standard errors in parentheses. Hh. observed ≥ 7years All households Mean diff. (std. err.) Mean (std. dev.) n Mean (std. dev.) n Household income (EUR) 2667.16 80,764 2646.61 183,870 20.56 (1413.96) (1419.32) (5.985) Food quantity (kg) 764.84 80,764 624.82 183,870 140.0 (430.38) (435.28) (1.831) BMI of female partner 25.09 76,876 24.92 173,027 0.172 (4.74) (4.85) (0.0209) Share of unhealthy food purchases 0.19 80,764 0.21 183,870 −0.0202 (0.11) (0.12) (0.000502) Household size 2.51 80,764 2.61 183,870 −0.0997 (1.32) (1.37) (0.00572) Age partner 1 53.04 80,764 47.66 183,870 5.377 (14.85) (15.33) (0.0641) Age partner 2 54.48 80,764 49.1 183,870 5.380 (15.02) (15.47) (0.0647) Partner 1 inactive = 1 0.41 80,764 0.32 183,870 0.0870 (0.49) (0.47) (0.00200) Partner 2 inactive = 1 0.37 80,764 0.27 183,870 0.0978 (0.48) (0.44) (0.00192) Household calorie needs 3417.47 77,600 3508.97 174,641 −91.50 (1797.38) (1801.73) (7.767) Meals at home per day 2.15 79,975 2.17 175,762 −0.0180 (1.12) (1.17) (0.00492) Broken homes andempty pantries: theimpact ofcouple separation… Page 31 of 56 26 Table 7 Evolution of the outcome variables per capita, per consumption unit and in levels around the time of separation (1) (2) (3) (4) (5) (6) (7) Per capita income Per capita food purchases Per cons. unit income Per cons. unit food purchases Income Food purchases BMI Event time –7 −0.0315 −0.0325 −0.0313 −0.0323 −60.19 20.62 0.147 (0.030) (0.048) (0.031) (0.050) (81.801) (37.087) (0.158) Event time –6 −0.00160 −0.0628 −0.0000452 −0.0613 −12.37 −1.627 0.0271 (0.019) (0.041) (0.019) (0.041) (50.562) (24.755) (0.104) Event time –5 0.00833 −0.0408 0.0127 −0.0364 16.02 16.53 0.0473 (0.017) (0.041) (0.016) (0.041) (42.330) (22.447) (0.090) Event time –4 −0.00339 −0.0434 −0.000629 −0.0407 −8.836 −3.417 0.0440 (0.015) (0.033) (0.014) (0.033) (41.073) (19.468) (0.093) Event time –3 −0.00348 −0.0211 0.000145 −0.0175 5.510 15.47 −0.0142 (0.014) (0.028) (0.014) (0.028) (35.718) (15.230) (0.068) Event time –2 0.00809 −0.0445 0.00943 −0.0431 21.48 0.158 0.0396 (0.008) (0.026) (0.008) (0.026) (20.076) (12.855) (0.048) Event time –1 0 0 0 0 0 0 0 (.) (.) (.) (.) (.) (.) (.) Event time 0 −0.0719*** 0.0289 −0.0727*** 0.0281 −156.0*** −33.70** −0.121** (0.010) (0.022) (0.010) (0.022) (24.885) (10.780) (0.040) Event time 1 −0.0883*** −0.103*** −0.105*** −0.120*** −482.7*** −113.2*** −0.345*** (0.016) (0.031) (0.015) (0.031) (37.774) (13.723) (0.058) Event time 2 −0.157*** −0.266*** −0.181*** −0.290*** −648.6*** −170.3*** −0.447*** (0.018) (0.039) (0.017) (0.039) (44.212) (15.618) (0.074) Event time 3 −0.169*** −0.258*** −0.196*** −0.285*** −682.4*** −165.8*** −0.322*** (0.021) (0.042) (0.021) (0.042) (53.712) (17.229) (0.091) J.Mink 26 Page 32 of 56 Table 7 (continued) (1) (2) (3) (4) (5) (6) (7) Per capita income Per capita food purchases Per cons. unit income Per cons. unit food purchases Income Food purchases BMI Event time 4 −0.153*** −0.282*** −0.180*** −0.309*** −597.2*** −153.0*** −0.0344 (0.027) (0.059) (0.027) (0.059) (83.288) (23.354) (0.128) Event time 5 −0.158*** −0.359*** −0.182*** −0.384*** −600.1*** −190.8*** −0.0340 (0.037) (0.079) (0.037) (0.079) (122.775) (29.289) (0.176) Event time 6 −0.135*** −0.185** −0.163*** −0.214*** −581.6*** −153.2*** 0.102 (0.038) (0.061) (0.038) (0.061) (119.209) (30.407) (0.210) Event time 7 −0.102* −0.0931 −0.135** −0.127 −477.0** −108.0** −0.0958 (0.050) (0.074) (0.051) (0.075) (164.296) (38.282) (0.244) Event time 8 −0.157* −0.0957 −0.197** −0.136 −644.7*** −141.5* −0.338 (0.076) (0.079) (0.073) (0.082) (156.443) (57.626) (0.345) Household size −0.295*** −0.198*** −0.227*** −0.129*** 145.5*** 114.3*** 0.0269 (0.004) (0.006) (0.004) (0.006) (11.383) (4.023) (0.020) Observations 80,860 80,860 80,860 80,860 80,860 80,860 77,111 R2 0.391 0.070 0.292 0.058 0.130 0.121 0.029 Year FE Yes Yes Yes Yes Yes Yes Yes Household FE Yes Yes Yes Yes Yes Yes Yes Age dummies Yes Yes Yes Yes Yes Yes Yes Robust standard errors clustered at the level of the household in parentheses. The table shows event time coefficients for the treatment group of households where the male partner leaves the female partner, relative to the control group of households where no separation occurs and relative to the year just before separation ( j =− 1 ). The dependent variables are per capital or per consumption unit disposable household income, food quantities purchased, and disposable household income, household food quantities purchased, the woman’s BMI measured in levels. In addition to household and year fixed effects and household size, all regressions include dummies for both partners’ ages. The data cover the period from 2005 to 2015. * p < 0.05 , ** p < 0.01 , *** p < 0.001 Broken homes andempty pantries: theimpact ofcouple separation… Page 33 of 56 26 Table 8 Evolution of outcomes around the time of separation - robustness to different model specifications Log income Log food Log BMI Share unhealthy food Event time -7 −0.00759 −0.00793 −0.00770 −0.0112 0.00830 0.00831 −0.0174** −0.0173** (0.032) (0.032) (0.050) (0.051) (0.006) (0.006) (0.006) (0.006) Event time -6 0.0153 0.0148 −0.0474 −0.0508 0.00322 0.00322 −0.0186*** −0.0185*** (0.019) (0.019) (0.042) (0.043) (0.004) (0.004) (0.005) (0.005) Event time -5 0.0201 0.0208 −0.0270 −0.0272 0.00297 0.00300 −0.0168*** −0.0166*** (0.016) (0.016) (0.041) (0.041) (0.003) (0.003) (0.005) (0.005) Event time -4 0.00342 0.00405 −0.0352 −0.0347 0.00159 0.00161 −0.0159*** −0.0158*** (0.015) (0.015) (0.033) (0.033) (0.003) (0.003) (0.004) (0.004) Event time -3 −0.00431 −0.00299 −0.0180 −0.0156 −0.000761 −0.000717 −0.0110*** −0.0109*** (0.014) (0.014) (0.029) (0.029) (0.002) (0.002) (0.003) (0.003) Event time -2 0.00623 0.00679 −0.0446 −0.0444 0.00133 0.00136 −0.00685** −0.00676** (0.008) (0.008) (0.026) (0.026) (0.002) (0.002) (0.002) (0.002) Event time -1 0 0 0 0 0 0 0 0 (.) (.) (.) (.) (.) (.) (.) (.) Event time 0 −0.0643*** −0.0651*** 0.0351 0.0348 −0.00480** −0.00484** 0.00179 0.00166 (0.010) (0.010) (0.022) (0.022) (0.002) (0.002) (0.002) (0.002) Event time 1 −0.188*** −0.180*** −0.189*** −0.188*** −0.0115*** −0.0110*** 0.00805* 0.00959** (0.014) (0.014) (0.031) (0.031) (0.002) (0.002) (0.003) (0.003) Event time 2 −0.261*** −0.254*** −0.356*** −0.358*** −0.0154*** −0.0150*** 0.00540 0.00681 (0.016) (0.016) (0.038) (0.038) (0.003) (0.003) (0.004) (0.004) Event time 3 −0.272*** −0.266*** −0.352*** −0.356*** −0.0108** −0.0104** 0.0000122 0.00131 (0.020) (0.020) (0.041) (0.041) (0.003) (0.003) (0.005) (0.005) Event time 4 −0.259*** −0.252*** −0.370*** −0.373*** −0.000617 −0.000210 0.00647 0.00774 (0.027) (0.027) (0.060) (0.060) (0.005) (0.005) (0.006) (0.006) J.Mink 26 Page 34 of 56 Table 8 (continued) Log income Log food Log BMI Share unhealthy food Event time 5 −0.276*** −0.268*** −0.476*** −0.474*** −0.00150 −0.00108 0.00598 0.00728 (0.039) (0.039) (0.078) (0.078) (0.007) (0.007) (0.008) (0.008) Event time 6 −0.271*** −0.263*** −0.332*** −0.329*** 0.00492 0.00536 −0.00502 −0.00365 (0.040) (0.040) (0.059) (0.059) (0.008) (0.008) (0.008) (0.008) Event time 7 −0.244*** −0.236*** −0.226** −0.224** −0.00206 −0.00163 −0.00728 −0.00595 (0.051) (0.051) (0.076) (0.076) (0.009) (0.009) (0.012) (0.012) Event time 8 −0.321*** −0.313*** −0.242** −0.242** −0.00901 −0.00855 −0.00988 −0.00843 (0.068) (0.067) (0.082) (0.084) (0.012) (0.012) (0.012) (0.012) Household size 0.0697*** 0.180*** 0.00137* 0.00413*** (0.004) (0.006) (0.001) (0.001) Consumption unit 0.0113*** 0.0263*** 0.000268* 0.000816*** (0.001) (0.001) (0.000) (0.000) Constant 7.660*** 7.611*** 12.79*** 12.73*** 3.210*** 3.208*** 0.184*** 0.178*** (0.009) (0.011) (0.017) (0.020) (0.002) (0.002) (0.002) (0.002) Observations 80,860 80,860 80,860 80,860 77,111 77,111 80,860 80,860 R2 0.131 0.133 0.081 0.078 0.021 0.021 0.035 0.035 Year FE Yes Yes Yes Yes Yes Yes Yes Yes Household FE Yes Yes Yes Yes Yes Yes Yes Yes Robust standard errors clustered at the level of the household in parentheses. The table shows event time coefficients for the treatment group of households where the male partner leaves the female partner, relative to the control group of households where no separation occurs and relative to the year just before separation ( j =− 1 ). The dependent variables are the logarithm of household disposable income, the logarithm of food quantities purchased, the logarithm of the woman’s BMI and the share of unhealthy food purchases in total food purchases. All regressions include household and year fixed effects. The data cover the period from 2005 to 2015. * p < 0.05 , ** p < 0.01 , *** p < 0.001 Broken homes andempty pantries: theimpact ofcouple separation… Page 35 of 56 26 Table 9 Evolution of meals eaten at home, female partner’s labor market status, and repartnering status around the time of separation Robust standard errors clustered at the level of the household in parentheses. The table shows event time coefficients for the treatment group of households where the male partner leaves the female partner, relative to the control group of households where no separation occurs and relative to the year just before separation ( j =− 1 ). The dependent variables are the average number of meals eaten at home in a typical week and the labour market status of the female partner who remains in the household after separation, measured as a dummy variable equal to one if she is inactive. Nb. meals Women is inactive = 1 Women repartnered = 1 Event time −7 0.0834 −0.00227 0.00210 (0.056) (0.024) (0.002) Event time −6 0.0175 −0.00466 0.00467** (0.030) (0.023) (0.002) Event time −5 0.0678* 0.00164 0.00400** (0.029) (0.018) (0.001) Event time −4 0.0306 −0.00663 0.00358** (0.024) (0.014) (0.001) Event time −3 0.00975 −0.0105 0.000606 (0.014) (0.011) (0.001) Event time −2 0.0249 −0.000990 −0.0000905 (0.013) (0.006) (0.001) Event time −1 0 0 0 (.) (.) (.) Event time 0 −0.0767*** −0.0155** −0.0000642 (0.011) (0.005) (0.000) Event time 1 0.225*** −0.0320*** 0.0000181 (0.019) (0.007) (0.000) Event time 2 0.0816*** −0.0428*** 0.0506*** (0.019) (0.009) (0.009) Event time 3 0.0571** −0.0299** 0.0296** (0.020) (0.012) (0.009) Event time 4 0.117* −0.0474*** 0.0281* (0.055) (0.014) (0.013) Event time 5 0.146 −0.0210 0.0104 (0.096) (0.026) (0.016) Event time 6 −0.0116 −0.0339 0.00684 (0.048) (0.029) (0.015) Event time 7 −0.0172 −0.0258 −0.00855 (0.060) (0.041) (0.007) Event time 8 −0.0782 −0.0753 −0.000133 (0.124) (0.039) (0.008) Constant 0.901*** 0.445*** −0.000192 (0.004) (0.003) (0.000) Observations 80,027 80,860 80,860 R2 0.010 0.023 0.032 Year fixed effects Yes Yes Yes Household fixed effects Yes Yes Yes J.Mink 26 Page 36 of 56 Fig. 4 Trends in outcome variables around separation for a all households, regardless of the sex of the partner who remains, and b households in which the male partner remains after separation. The figure shows event time coefficients relative to the control group of households where no separation occurs and relative to the year just before separation ( j =− 1 ). a Results for all households, while b shows results for households where the remaining partner is male. The dependent variables are the logarithm of income, food quantities purchased, remaining partner’s BMI, and the share of unhealthy food products purchased by household i in year t. In addition to household and year fixed effects and household size, all regressions include dummies for both partners’ ages. The data cover the period from 2005 to 2015. The 95% confidence intervals are based on standard errors clustered at the level of the household ▸ Broken homes andempty pantries: theimpact ofcouple separation… Page 37 of 56 26 J.Mink 26 Page 38 of 56 Fig. 5 Trends in the probability of being underweight, normal weight, overweight or obese around separation. The figure shows event time coefficients for the treatment group of households where the male partner leaves the female partner, relative to the control group of households where no separation occurs and relative to the year just before separation ( j =− 1 ). The dependent variables are dummy variables indicating whether the female partner in household i in year t is underweight (variable equals 1 for a BMI < 18.5 and 0 otherwise), of normal weight (variable equals 1 for a BMI ≥ 18.5 and < 25 and 0 otherwise), overweight (variable equals 1 for a BMI ≥ 25 and < 30 and 0 otherwise) or obese (variable equals 1 for a BMI > 30 and 0 otherwise). In addition to household and year fixed effects and household size, all regressions include dummies for both partners’ ages. The data cover the period from 2005 to 2015. The 95% confidence intervals are based on standard errors clustered at the level of the household Broken homes andempty pantries: theimpact ofcouple separation… Page 39 of 56 26 Fig. 6 Trend in outcome variables around separation using per capita or per consumption unit income and food purchases. The figure shows event time coefficients for the treatment group of households where the male partner leaves the female partner, relative to the control group of households where no separation occurs and relative to the year just before separation ( j =− 1 ). The dependent variables are the per capita or per consumption unit income, food quantities purchased by household i in year j. Consumption units are constructed by assigning a weight of 1 to the first adult, 0.7 to each additional person aged 15 or older, and 0.5 to each child under 15, to reflect age-specific needs and economies of scale. In addition to household and year fixed effects and household size, all regressions include dummies for both partners’ ages. The data cover the period from 2005 to 2015. The 95% confidence intervals are based on standard errors clustered at the level of the household J.Mink 26 Page 40 of 56 Fig. 7 Trend in outcome variables around separation using the original non-transformed dependent variables. The figure shows event time coefficients for the treatment group of households where the male partner leaves the female partner, relative to the control group of households where no separation occurs, and relative to the year just before separation ( j =− 1 ). The dependent variables are the income, food quantities purchased, remaining partner’s BMI, the share of unhealthy food products purchased in total quantities purchased by household i in year t. In addition to household and year fixed effects and household size, all regressions include dummies for both partners’ ages. The data cover the period from 2005 to 2015. The 95% confidence intervals are based on standard errors clustered at the level of the household Broken homes andempty pantries: theimpact ofcouple separation… Page 47 of 56 26 Fig. 14 Trend in outcome variables around placebo dates of separation. The figure shows the mean and the 95th percentile of the coefficient distributions of placebo exercises, that is, event time coefficients estimated relative to the placebo control group of households and relative to the year just before the placebo date of separation ( j =− 1 ). The upper four graphs show the effects of assigning random separation dates in households that do not separate, while the lower four graphs show the effects of randomly reassigning separation dates in the households that separate. The dependent variables are the logarithm of income, food quantities purchased, female partner’s BMI, and the share of unhealthy food products purchased by household i in year t. In addition to household and year fixed effects and household size, all regressions include dummies for both partners’ ages. The data cover the period from 2005 to 2015. The 95% confidence intervals are based on standard errors clustered at the level of the household J.Mink 26 Page 48 of 56 Fig. 15 Trend in the probability of being underweight, normal weight or overweight or obese, by household income. The figure shows event time coefficients for the treatment group of households where the male partner leaves the female partner, relative to the control group of households where no separation occurs and relative to the year just before separation ( j =− 1 ). The left-hand panels present results for the 30% of households with the lowest pre-separation per capita incomes, while the right-hand panels show results for the 30% of households with the highest incomes. The dependent variables are dummy variables indicating whether the female partner in household i in year t is underweight (variable equals 1 for a BMI < 18.5 and 0 otherwise), of normal weight (variable equals 1 for a BMI ≥ 18.5 and < 25 and 0 otherwise), overweight (variable equals 1 for a BMI ≥ 25 and < 30 and 0 otherwise), or obese (variable equals 1 for a BMI > 30 and 0 otherwise). In addition to household and year fixed effects and household size, all regressions include dummies for both partners’ ages. The data cover the period from 2005 to 2015. The 95% confidence intervals are based on standard errors clustered at the level of the household ▸ Broken homes andempty pantries: theimpact ofcouple separation… Page 49 of 56 26 Difference relative to j -1 Years since break-up 30% poorest households Difference relative to j -1 Years since break-up 30% richest households A. Underweight Difference relative to j -1 Years since break-up 30% poorest households Difference relative to j -1 Years since break-up 30% richest households B. Normal weight Difference relative to j -1 Years since break-up 30% poorest households Difference relative to j -1 Years since break-up 30% richest households C. Overweight or obese J.Mink 26 Page 50 of 56 Fig. 17 Trend in outcome variables around separation, by pre-separation employment status of the female partner. The figure shows event time coefficients for the treatment group of households where the male partner leaves the female partner, relative to the control group of households where no separation occurs and relative to the year just before separation ( j =− 1 ). The left-hand panel presents results for households where the woman is employed and the right-hand panel where the woman is not employed. The dependent variables are the logarithm of income, food quantities purchased, female partner’s BMI, and the share of unhealthy food products purchased by household i in year t. In addition to household and year fixed effects and household size, all regressions include dummies for both partners’ ages. The data cover the period from 2005 to 2015. The 95% confidence intervals are based on standard errors clustered at the level of the household ▸ Difference relative to t -1 Years since break-up 30% poorest households, couples only households Difference relative to t -1 Years since break-up 30% richest households, couples only households Difference relative to t -1 Years since break-up 30% poorest households, households with children Difference relative to t -1 Years since break-up 30% richest households, households with childre n A. Income Difference relative to t -1 Years since break-up 30% poorest households, couples only households Difference relative to t -1 Years since break-up 30% richest households, couples only households Difference relative to t -1 Years since break-up 30% poorest households, households with children Difference relative to t -1 Years since break-up 30% richest households, households with children B. Food quantities purchased Difference relative to t -1 Years since break-up 30% poorest households, couples only households Difference relative to t -1 Years since break-up 30% richest households, couples only households Difference relative to t -1 Years since break-up 30% poorest households, households with children Difference relative to t -1 Years since break-up 30% richest households, households with children C. Spouse's body weight index Difference relative to t -1 Years since break-u p 30% poorest households, couples only households Difference relative to t -1 Years since break-u p 30% richest households, couples only households Difference relative to t -1 Years since break-u p 30% poorest households, households with children Difference relative to t -1 Years since break-u p 30% richest households, households with children D. Share of unhealthy food purchased Fig. 16 Trend in outcome variables around separation, by family composition and pre-separation income. The figure shows event time coefficients for the treatment group of households where the male partner leaves the female partner, relative to the control group of households where no separation occurs and relative to the year just before separation ( j =− 1 ). The panels show the results of separate regressions for household groups defined by pre-separation household income and family composition (couple-only households or households with children under 18 at the time of the separation). The dependent variables are the logarithm of income, food quantities purchased, female partner’s BMI, and the share of unhealthy food products in total food purchased by household i in year t. In addition to household and year fixed effects and household size, all regressions include dummies for both partners’ ages. The data cover the period from 2005 to 2015. The 95% confidence intervals are based on standard errors clustered at the level of the household Broken homes andempty pantries: theimpact ofcouple separation… Page 51 of 56 26 Percentage difference relative to t -1 Years since break-up Spouse is active when separating Percentage difference relative to t -1 Years since break-up Spouse is inactive when separating A. Income Percentage difference relative to t -1 Years since break-up Spouse is active when separating Percentage difference relative to t -1 Years since break-up Spouse is inactive when separating B. Food quantities purchased Percentage difference relative to t -1 Years since break-up Spouse is active when separating Percentage difference relative to t -1 Years since break-up Spouse is inactive when separating C. Spouse's body weight index Difference relative to t -1 Years since break-up Spouse is active when separating Difference relative to t -1 Years since break-up Spouse is inactive when separating D. Share of unhealthy food purchased J.Mink 26 Page 52 of 56 Fig. 18 Trend in outcome variables around separation, robustness to unobserved effect heterogeneity. The figure shows coefficients estimated using the approach proposed by De Chaisemartin and d’Haultfoeuille (2020) for estimates that are robust to unobserved effect heterogeneity (implemented via the did_multiplegt package). The approach relies on not-yet-treated units and the parallel trends assumption to identify treatment effects for each treated unit type, which can then be averaged together. The dependent variables are the logarithm of income, food quantities purchased, female partner’s BMI, and the share of unhealthy food products purchased by household i in year t. In addition to household and year fixed effects and household size, all regressions include dummies for both partners’ ages. The data cover the period from 2005 to 2015. The 95% confidence intervals are based on standard errors clustered at the level of the household Broken homes andempty pantries: theimpact ofcouple separation… Page 53 of 56 26 Fig. 19 Trend in outcome variables around separation for the subgroup of households in which the woman becomes active in the labour market after separation. The figure shows event time coefficients for the treatment group of households where the male partner leaves the female partner, relative to the control group of households where no separation occurs and relative to the year just before separation ( j =− 1 ). The figure presents results for the subgroup of households in which the remaining partner (the woman) becomes employed after the separation. The dependent variables are the logarithm of income, food quantities purchased, the woman’s BMI, and the share of unhealthy food products in total food purchased by household i in year t. In addition to household and year fixed effects and household size, all regressions include dummies for both partners’ ages. The data cover the period from 2005 to 2015. The 95% confidence intervals are based on standard errors clustered at the level of the household J.Mink 26 Page 54 of 56 Percentage difference relative to t -1 Years since break-up A . Income Percentage difference relative to t -1 Years since break-up B. Foo d quantities purc h ase d Percentage difference relative to t -1 Years since break-up C. Spouse's body weight index Difference relative to t -1 Years since break-up D. Share of unhealthy food purchased Fig. 20 Trend in outcome variables around separation for the subgroup of households in which the woman enters a new relationship after the separation. The figure shows event time coefficients for the treatment group of households where the male partner leaves the female partner, relative to the control group of households where no separation occurs and relative to the year just before separation ( j =− 1 ). The figure shows results for the subgroup of households in which the women enters a new relationship after the separation. The dependent variables are the logarithm of income, food quantities purchased, woman’s BMI, and the share of unhealthy food products purchased by household i in year t. In addition to household and year fixed effects and household size, all regressions include dummies for both partners’ ages. The data cover the period from 2005 to 2015. The 95% confidence intervals are based on standard errors clustered at the level of the household Acknowledgement This research is supported by the Strategic Research Initiative NutriPerso from University Paris-Saclay and the project AlimaSSen [ANR-14-CE20-0003-01] from the French National Research Agency (ANR). Support by the German Research Foundation (DFG) through CRC TR 224 (Project B07) and through Germany’s Excellence Strategy (EXC 2126/1-390838866) is gratefully acknowledged. Author contributions JM made substantial contributions to the conception of the work, the analysis and interpretation of data; JM drafted the work; JM approved the version to be published and agrees to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. Funding Open Access funding enabled and organized by Projekt DEAL. Broken homes andempty pantries: theimpact ofcouple separation… Page 55 of 56 26 Data availability The data cannot be made publicly available by the author as this study uses data from a proprietary consumer panel which the author is not authorised to share. The author will provide information on how to obtain the data from Kantar Worldpanel upon request. Declarations Conflict of interest The authors declare no conflict of interest. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/ licenses/by/4.0/. References Avellar S, Smock PJ (2005) The economic consequences of the dissolution of cohabiting unions. J Marriage Fam 67(2):315–327 Bianchi SM, McArthur E (1991) Family disruption and economic hardship: the short-run picture for children. US Bureau of Census, Current Population Report Series, p 70 Bonnet C, Garbinti B, Solaz A, etal (2015) Les conditions de vie des enfants aprãs le divorce. Technical report Bonnet C, Garbinti B, Solaz A (2021) The flip side of marital specialization: the gendered effect of divorce on living standards and labor supply. J Popul Econ 34:515–573 Costemalle V (2017) Formations et ruptures d’unions: quelles sont les spécificités des unions libres? INSEE références-France portrait social, pp 95–110 De Chaisemartin C, d’Haultfoeuille X (2020) Two-way fixed effects estimators with heterogeneous treatment effects. Am Econ Rev 110(9):2964–2996 de Saint Pol T (2006) Corps et appartenance sociale: la corpulence en europe. Données sociales 2006-La société française, pp 649–656 De Vaus D, Gray M, Qu L, Stanton D (2014) The economic consequences of divorce in Australia. Int J Law Policy Fam 28(1):26–47 Duncan GJ, Hoffman SD (1985a) Economic consequences of marital instability. In: Horizontal equity, uncertainty, and economic well-being, University of Chicago Press, pp 427–470 Duncan GJ, Hoffman SD (1985b) A reconsideration of the economic consequences of marital dissolution. Demography 22(4):485–497 Eng PM, Kawachi I, Fitzmaurice G, Rimm EB (2005) Effects of marital transitions on changes in dietary and other health behaviours in us male health professionals. J Epidemiol Commun Health 59(1):56–62 Fisher H, Low H (2009) Sharing lives, dividing assets: an interdisciplinary study. Hart Publishing Fisher H, Low H (2016) Recovery from divorce: comparing high and low income couples. Int J Law Policy Fam 30(3):338–371 Galarneau D, Sturrock J (1997) Family income after separation. Statistics Canada Hoffman S (1977) Marital instability and the economic status of women. Demography 14(1):67–76 Holden KC, Smock PJ (1991) The economic costs of marital dissolution: Why do women bear a disproportionate cost? Ann Rev Sociol 17(1):51–78 INED (2018) Mortality in France. https:// www. ined. fr/ fr/ toutsavoirpopul ation/ chiff res/ france/ morta litecausedeces/ tablemorta lite INSEE (2015) Couples et families. https:// www. insee. fr/ fr/ stati stiqu es/ 20175 10? somma ire= 20175 28 INSEE (2016) 123 500 divorces en 2014 J.Mink 26 Page 56 of 56 Lee S, Cho E, Grodstein F, Kawachi I, Hu FB, Colditz GA (2004) Effects of marital transitions on changes in dietary and other health behaviours in us women. Int J Epidemiol 34(1):69–78 Mata J, Richter D, Schneider T, Hertwig R (2018) How cohabitation, marriage, separation, and divorce influence BMI: a prospective panel study. Health Psychol 37(10):948 McKeever M, Wolfinger NH (2001) Reexamining the economic costs of marital disruption for women. Soc Sci Q 82(1):202–217 McLanahan S, Sandefur G (1994) Growing up with a single parent. What Hurts, What Helps. ERIC McLanahan S, Tach L, Schneider D (2013) The causal effects of father absence. Ann Rev Sociol 39:399–427 McManus PA, DiPrete TA (2001) Losers and winners: the financial consequences of separation and divorce for men. Am Sociol Rev 66(2):246 Miller DL (2023) An introductory guide to event study models. J Econ Perspect 37(2):203–230 OECD (2011) Doing better for families OECD (2017) Public spending on family benefits. https:// www. oecd. org/ els/ family/ datab ase. htm Page ME, Stevens AH (2004) The economic consequences of absent parents. J Hum Resour 39(1):80–107 Peterson RR (1996) A re-evaluation of the economic consequences of divorce. Am Sociol Rev 61:528–536 Smock PJ (1994) Gender and the short-run economic consequences of marital disruption. Soc Forces 73(1):243–262 Sun L, Abraham S (2021) Estimating dynamic treatment effects in event studies with heterogeneous treatment effects. J Econom 225(2):175–199 Tach LM, Eads A (2015) Trends in the economic consequences of marital and cohabitation dissolution in the united states. Demography 52(2):401–432 Vinther JL, Conklin AI, Wareham NJ, Monsivais P (2016) Marital transitions and associated changes in fruit and vegetable intake: findings from the population-based prospective epic-norfolk cohort, uk. Soc Sci Med 157:120–126 Weiss RS (1984) The impact of marital dissolution on income and consumption in single-parent households. J Marriage Fam 46:115–127 Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.