A family affair? Long-term economic and mental health effects of spousal cancer

Böckerman, Petri; Kortelainen, Mika; Salokangas, Henri; Vaalavuo, Maria Article — Published Version A family affair? Long-term economic and mental health effects of spousal cancer Journal of Population Economics Provided in Cooperation with: Springer Nature Suggested Citation: Böckerman, Petri; Kortelainen, Mika; Salokangas, Henri; Vaalavuo, Maria (2025) : A family affair? Long-term economic and mental health effects of spousal cancer, Journal of Population Economics, ISSN 1432-1475, Springer, Berlin, Heidelberg, Vol. 38, Iss. 1, https://doi.org/10.1007/s00148-025-01070-x This Version is available at: https://hdl.handle.net/10419/318427 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/ Journal of Population Economics (2025) 38:19 https://doi.org/10.1007/s00148-025-01070-x ORIGINAL PAPER A family affair? Long-term economic and mental health effects of spousal cancer Petri Böckerman1,2,3 ·Mika Kortelainen4,5 ·Henri Salokangas5· Maria Vaalavuo5 Received: 19 March 2023 / Accepted: 2 January 2025 © The Author(s) 2025 Abstract Research on the family spillover effects of health shocks, which has focused mainly on labor market outcomes, has yielded inconclusive results, with limited insight into long-term consequences or underlying mechanisms. We analyze the shortand longterm impacts of cancer on the unaffected spouse’s labor supply and mental health as well as marital stability, considering gender and relative income status within the households. Using population-based register data from Finland (1995–2019) and a dynamic difference-in-differences design, we observe two key findings. First, a cancer diagnosis leads to very modest changes in a spouse’s labor supply but significant increases in the likelihood of psychotropic drug use and psychiatric outpatient visits. Second, the main results mask considerable heterogeneity regarding relative income within the household. Secondary earners increase their labor supply in response to fatal cancers but decrease it in non-fatal cases, while breadwinners show small negative responses in both. Bereaved women with lower income share experience more psychiatric symptoms, a trend not observed in men. Our findings reveal the importance of pre-shock breadwinner status in family responses to health shocks, suggesting the need for targeted support for caregiving and bereaved spouses. Keywords Health shock ·Cancer ·Family spillover effects ·Household division of labor ·Event study ·Difference-in-differences ·Mental health ·Marital stability JEL Classification I10 ·J12 ·J17 ·J22 1 Introduction There is a growing body of research on the indirect effects of severe health shocks within the family (e.g., García-Gómez et al. 2013; Jeon and Pohl 2017; Fadlon and Nielsen 2021; Vaalavuo et al. 2023). Indirect effects pertain to family members other Responsible editor:XiChen Extended author information available on the last page of the article 0123456789().: V,-vol 123 19 Page 2 of 30 P. Böckerman et al. than the person who falls ill. Accurately quantifying such effects, which have been largely neglected in the evaluation of the cost-effectiveness of healthcare services, can provide novel insights for designing health interventions and deepen our understanding of the interconnections between health and labor market outcomes (Böckerman and Ilmakunnas 2009; Picchio and Ubaldi 2024). This perspective is particularly relevant for cancers, which are a major contributor to the global disease burden, with their impact expected to increase in the coming decades due to aging populations (Global Burden of Disease 2019 Cancer Collaboration 2022). Therefore, the wellbeing impacts of cancer are policy-relevant globally. There are many possible reasons for indirect effects. First, the health shock may cause the person falling ill to reduce their effective labor supply or even withdraw completely from the labor market. It has previously been demonstrated that a cancer diagnosis can lead to significant economic losses in households across various institutional contexts (Bradley et al. 2005; Jeon 2017; Vaalavuo 2021). This can disturb the pre-existing arrangement on the joint labor supply of spouses, especially when the person falling ill is the main breadwinner in the family.1Consequently, the unaffected spouse may react to this by increasing her/his labor supply to maintain the family’s income level and material well-being, in the spirit of what has often been called the “added worker hypothesis” (Mincer 1962) or “added worker effect” (Lundberg 1985). Limited social security, substantial debt, restricted access to credit, and high healthcare costs can significantly exacerbate this effect. The second potential driver of the indirect labor market effects leads to the opposite consequences. The healthy spouse might reduce her/his labor supply due to the provision of care (known as the “caregiver effect”) or concerns for the physical/mental well-being of a close family member and a desire for more shared leisure time (the “family effect”).2 Beyond affecting labor supply, a cancer diagnosis can cause psychological distress among family members. The uncertainty about recurrence, survival, and returning to work can significantly impact not just the patient but also their loved ones, leading to lasting effects on their lives (Mellon et al. 2007; Guan et al. 2021). This distress may extend to healthy spouses as well, as some evidence suggests (Hu et al. 2023; Angelini and Costa-Font 2023). Moreover, cancer, as a sudden, severe, and unexpected shock to the family, can significantly affect the marital stability of couples navigating this difficult situation (Syse 2008). While family spillover effects have recently attracted increased interest among researchers and policy advocates in the field, empirical evidence on the topic remains inconclusive. The divergent results of existing research are likely explained, in part, by differences in institutional contexts, data characteristics, empirical methods applied, and the specific health shocks examined. To further complicate the picture, existing patterns in the division of household labor supply may substantially affect the response of the unaffected spouse (see also Riekhoff and Vaalavuo 2021; Vaalavuo et al. 2023). The couple’s joint pattern of labor supply prior to the health shock affects the need, opportunities, and obstacles to adjusting labor market participation. In addition to labor 1Throughout the paper “spouse” refers to a significant other in a marriage or cohabitation. 2For a detailed discussion of the relevant terminology, see Bobinac et al. (2010). 123 A family affair? Long-term economic and mental health… Page 3 of 30 19 market responses, these features may also influence other dimensions of well-being, including psychological health and marital stability. In this article, our principal research question is how a spouse’s cancer affects the labor market outcomes and mental health of the healthy spouse and the couple’s marital stability. Furthermore, while labor market responses have been reported in the prior literature, we contribute by investigating the heterogeneous responses by relative income position within the household. Identifying the causal effects of a spouse’s cancer diagnosis is challenging. Our identification strategy relies on quasi-random variation in the timing of the cancer diagnoses within the estimation window, using a dynamic difference-in-differences or event study design, as outlined by Fadlon and Nielsen (2019,2021). This approach also enables us to assess the key identification assumptions of the empirical specification. Our study advances the understanding of cancer’s broader impacts on well-being in three important ways. First, we integrate theories and concepts of household division of labor into the empirical models. We examine the heterogeneous impacts by the relative income status of the spouses prior to the cancer diagnosis separately for men and women. Relative income is relevant because it entails information on the potential financial losses for the surviving spouse caused by cancer, and signals potential economic independence within the household. While some previous studies have examined gender differences in the spillover effects, they have not considered the effect of relative income status separately (independently of gender) or focused on long-term impacts. Second, our study focuses on the psychological spillover effects of cancer and the impacts on marital stability. These two dimensions of well-being have been only rarely examined in the literature, especially in connection to relative income status. Third, in contrast to many other studies, we use particularly long follow-up time as some attributes associated with cancer, e.g., uncertainty about recurrence, years of survival, and return to work, may have a prolonged effect on the cancer patients’ lives but also on the lives of their loved ones. Using the panel structure of our data from Finland over the years 1995–2019, we follow couples 5 years before and 10 years after the initial cancer diagnosis. While there exists only a nascent body of quasi-experimental research on the indirect effects of health shocks on the spouse’s labor market outcomes, the connection between health and labor supply within the family is not new to the field of economics. Already in the 1970s, Parsons (1977) analyzed the impact of family structure on men’s health, work hours, and earnings and observed that poor health significantly reduced men’s work hours and earnings but did not notably increase work hours among other family members. Berger (1983) and Berger and Fleisher (1984) also found only small increases in wives’ work hours and no substantial impact on labor force participation in response to husbands’ poor health. In a more recent and more closely related study, Jeon and Pohl (2017) examined the effect of different cancer diagnoses on the spouse’s employment and earnings trajectories based on Canadian register data. We complement their study by analyzing the role of relative income within the household and by examining a wider set of outcomes, including psychotropic drug use. We are not aware of any previous quasiexperimental studies analyzing the spillover effects of a health shock on mental health. 123 19 Page 4 of 30 P. Böckerman et al. While some studies (e.g., Bom et al. 2019; Stöckel and Bom 2022; Angelini and Costa-Font 2023) have examined mental health outcomes among spouses, they have not employed dynamic difference-in-differences or similar identification strategies to tease out causal effects. Methodologically, the closest to our study is Fadlon and Nielsen (2021), who investigated households’ labor supply responses to fatal and severe non-fatal health shocks using Danish data. While using a similar identification strategy, we concentrate on a different health shock (i.e., cancer) and provide evidence also on the psychological well-being effects. Moreover, we analyze the heterogeneity of labor supply responses based on the relative income position of couples and evaluate spillover effects both in the short and long run. Our study is also related to concurrent work by Arrieta and Li (2023) that focuses on the effect of emergency department visits (i.e., acute health shocks requiring urgent care but with a potentially short duration) on intra-family adjustment of labor supply and care in the U.S. context. Overall, we contribute to the emerging literature on the topic by investigating longer-term impacts on labor supply and mental well-being as well as the potential mechanisms behind the spousal effects in more detail. Our findings from Finland, a comprehensive Nordic welfare state, are likely to illuminate the institutional differences that drive labor supply responses in various country contexts. Additionally, the Finnish context holds broader interest for two other reasons. First, we examine the effects of cancer on total family income, including received social transfers. This issue is highly policy-relevant in other high-income countries as they develop more comprehensive social safety nets for families to tackle the financial burden caused by chronic illnesses. Second, Finland’s cancer survival rates are among the world’s highest, which highlights the importance of understanding the indirect effects of a health shock at the family level. In the near future, the indirect labor market effects of poor health might become particularly salient in aging societies that aim to prolong working careers while reconciling informal care and paid work among older employees. We observe that female spouses increase their employment for some years after a severe health shock, which is consistent with the added worker effect. However, the magnitude of the impact on annual earnings is negligible. Among men, we observe the opposite: male spouses’ earnings decrease once their partner falls ill. Overall, cancer diagnosis causes rather small changes in the labor supply of spouses but relatively large increases in the use of psychotropic medication and psychiatric outpatient visits. More importantly, our results shed light on labor supply responses within families, considering both the breadwinner status and the survival of the cancer patient. We observe that both men and women experience decreasing earnings in cases of nonfatal cancers, with greater deficits for secondary earners. However, responses in the extensive margin are negligible. Both women and men moderately increase their psychotropic medication, with the most significant rise observed among secondary-earner women. Notably, secondary-earner women also experience a positive effect on marital stability, while men and breadwinner women remain unaffected. In fatal cancers, a clearer relationship emerges between relative income and labor supply decisions. The earnings responses are linearly related to the pre-cancer income share within 123 A family affair? Long-term economic and mental health… Page 5 of 30 19 the household, indicating that the greater the income share of the deceased spouse, the greater the increase in the labor supply at the intensive margin of the surviving spouse. In the long-term, secondary-earner women also demonstrate increased labor supply at an extensive margin. This implies that the surviving spouse compensates for the economic loss by adding labor supply. At the same time, this group shows the most substantial increase in psychotropic medication use, suggesting a connection between the markedly increased labor supply and psychiatric symptoms from bereavement. Overall, the adverse mental health effects are substantial both among men and women, in both the shortand long-term. These findings highlight the importance of the follow-up duration, the survival status of the ill spouse, and the breadwinner status within the household in influencing the results. Consequently, they may help to resolve some inconsistencies present in the existing evidence on the topic. The article is structured as follows. Section 2offers an overview of the relevant literature. Section 3describes the Finnish register data and the empirical framework. Section 4presents the estimation results. The final section offers a comprehensive discussion of the key findings. 2 Conceptual framework Individuals consider the well-being and economic prospects of their entire household, not just their own, when making labor supply decisions (Mincer 1962; Blundell and Walker 1982; Becker 1991). Early in their relationship, spouses often negotiate the division of household labor, where traditionally, the husband specializes in paid work outside the home, and the wife, in unpaid household work at home (Becker 1991; Leira 1992). The household operates as an economic unit, sharing resources and risks. A health shock, like cancer, can significantly change the household’s economic situation, impacting labor supply, marital stability, and mental health, all crucial for assessing family well-being. Existing research on the spillover effects of health shocks on spouses has primarily studied labor market impacts, yielding inconclusive results. Notably, Coile (2004) examined heart attacks and new cancer diagnoses among older adults in the U.S., revealing only a small added worker effect for men and none for women. On the other hand, Jeon and Pohl (2017) observed a significant decrease in labor supply among Canadians whose spouses were diagnosed with cancer, particularly among men at the intensive margin. The authors interpret this finding as individuals reducing effective labor supply to provide care for their sick spouses and to share leisure time. Similarly, Anand et al. (2022) found a reduction in labor force participation among potential caregivers following a spouse’s health shock. In contrast, studies like Giaquinto et al. (2022) using UK data and Jolly and Theodoropoulos (2023) with SHARE data from Europe, show minimal changes in labor supply but highlight an increased focus on caregiving and a higher likelihood of retirement. Evidence on the long-term spillover effects of health shocks on psychological well-being is sparse, and dynamic difference-in-differences designs have been rarely utilized in this research. Recent studies using survival analysis and register data from 123 19 Page 6 of 30 P. Böckerman et al. Denmark and Sweden indicate that a spouse’s cancer diagnosis increases the risk of receiving a psychiatric diagnosis in hospital-based inpatient or outpatient care compared to matched controls (Hu et al. 2023). Survey evidence also supports the view that cancer elevates the psychological distress of spouses. This increased distress may stem from a greater caregiving burden, impacting the spouse’s mental health in a dose-response manner (Bom et al. 2019; Stöckel and Bom 2022), but also because the lives of spouses are intimately linked in terms of emotional well-being and family responsibilities (Northouse and McCorkle 2015). In a recent study closely related to ours, Angelini and Costa-Font (2023)usea cross-sectional survey data from Europe (SHARE) and find that fatal cancer is significantly associated with the surviving partner’s well-being, leading to increased depression, loneliness, and sleep problems. While being an important addition to the health spillover literature, their study departs from the conventional approach by using baseline characteristics and a first differences specification that may not fully account for unobserved factors influencing changes over time. Our article contributes to previous research by using more comprehensive register data on health-related consumption of psychotropic medicine and public psychiatric services within a dynamic difference-in-differences framework. Gender differences in spousal responses to health shocks, particularly cancer, are shaped by survival rates and the role of the primary breadwinner, often the male. The relative income of each spouse has a significant impact on how households respond to such shocks. If the primary breadwinner falls ill, significant labor adjustments may be needed, potentially leading to increased psychological stress and changes in family dynamics (Becker et al. 1977). However, when the secondary-earner spouse is affected, labor adjustments might be smaller. Consistent with this view, Fadlon and Nielsen (2021) found that surviving widows, but not widowers, increased their labor supply following fatal events, linking financial loss to labor force participation. This finding indicates that self-insurance might play a crucial role in how families adjust their labor in response to health shocks. Health shocks can affect marital stability, especially against the backdrop of changing gender norms in Nordic countries, where women increasingly academically outperform men, challenging the traditional male breadwinner model. Studies like Bertrand et al. (2015) show that female breadwinning can influence marital satisfaction and stability, a trend noted in the U.S. in the 1960s and 1970s but less so in the 1990s (Schwartz and Gonalons-Pons 2016), and is associated with increased marital dissolution (Foster and Stratton 2021). This highlights the need to explore how gender roles and relative household income influence marital stability following health shocks. Contrary to Becker et al. (1977), who suggested that health issues could negatively impact marital stability by affecting traits like income potential and health, recent studies offer different views. For example, Bünnings et al. (2021), analyzing German data, discovered that a spouse’s health decline does not necessarily lead to marital instability and may even strengthen the relationship. Moreover, Ehlert (2021) examined how a health shock’s impact on marital stability varies depending on the expected survivor’s pension, revealing a positive correlation between potential survivor benefits and the 123 A family affair? Long-term economic and mental health… Page 7 of 30 19 likelihood of staying married after a health shock for female partners. This result suggests that higher economic dependency on the affected spouse might enhance marital stability, as the unaffected partner may have fewer alternatives outside the marriage. Conversely, lower economic dependence could have the opposite effect. The existing literature provides insights into how the economic contributions of a partner, relative to those of the affected spouse, influence adjustments in labor supply and marital stability. However, the effects on mental health remain unexplored. Our study is the first to examine the role of relative income in moderating the mental health effects of a spouse’s health shock. We examine two hypotheses: First, in households where the unaffected spouse contributes less economically, they may face greater mental health challenges due to financial stress. Second, if the unaffected spouse is the primary earner, they might encounter less financial stress but more emotional and caregiving burdens, impacting their mental health differently. 3 Empirical approach 3.1 Research design and identification Our empirical approach employs the dynamic difference-in-differences design, similar to Fadlon and Nielsen (2019,2021), hereafter referred to as FN DiD. By utilizing this identification strategy, we created counterfactual scenarios for couples in which one spouse received a cancer diagnosis. These were derived from among couples who were diagnosed years later. Employing households affected by cancer as the control group is designed to reduce the selection bias encountered in straightforward case-control comparisons. Figure 1illustrates our research design using examples where the treatment groups consist of individuals whose spouses were diagnosed with cancer in the year 2000. In Panel A, we plot the yearly indicator for any psychotropic drug purchase for this treatment group and compare it to the trajectory of the same outcome in individuals whose spouses were diagnosed with a cancer diagnosis 11 years later (=11). The follow-up continues until the last year when the control group has not yet been treated, i.e., year 2010. Although the pre-event levels of the two groups are very similar, direct comparison may not be appropriate due to differences in age and sex distributions between the groups. To ensure a valid comparison, we weighted the outcomes in the control groups according to the age-sex distribution of the treatment group, effectively mimicking matching along these dimensions. Panel B demonstrates that weighting slightly alters the outcome level for the control group once age and sex are adjusted. Both Panels A and B indicate that the likelihood of psychotropic drug purchases significantly increases for individuals once their spouse is diagnosed with cancer, compared to the control group. In Panel C, we supplement the comparison with two alternative control groups: those affected by a spouse’s cancer diagnosis in 2006 (=6) and those whose spouse have not been diagnosed with cancer. The outcome dynamics and levels in these alternative control groups are very similar until 2006. Afterward, the group diagnosed with cancer 123 19 Page 8 of 30 P. Böckerman et al. A. Research design with , unweighted 2000 2011 .06 .08 .1 .12 .14 .16 .18 .2 Psychotropic medication (pp.) 1995 2000 2005 2010 Year B. Research design with , weighted 2000 2011 .06 .08 .1 .12 .14 .16 .18 .2 Psychotropic medication (pp.) 1995 2000 2005 2010 Year C. Comparison with alternative control groups No diagnosis 2000 2006 2011 .06 .08 .1 .12 .14 .16 .18 .2 Psychotropic medication (pp.) 1995 2000 2005 2010 Year D. Construction of counterfactual 2011 2000 Counterfactual .06 .08 .1 .12 .14 .16 .18 .2 Psychotropic medication (pp.) -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 Years since cancer diagnosis Fig. 1 Illustrations of the FN DiD design with psychotropic medication purchases. Panel A compares the treatment group with cohorts that received a cancer diagnosis 11 years later. Panel B includes the responses of households that underwent the same shock in different years, alongside matched controls that have no history of cancer diagnosis. Panel C demonstrates the construction of the counterfactual, based on the precancer difference in outcomes. The outcomes for potential control groups are adjusted based on the age and sex distribution of the treatment group in 2006 becomes “treated” and is no longer viable as a control group for studying the longer-term impacts of cancer. The focus on the economic impacts of cancer typically spans 3–5 years in related studies (e.g., Jeon 2017; Jeon and Pohl 2017,2019; Vaalavuo 2021). However, Tai et al. (2005) argue that for certain cancers, like pancreatic and stomach cancers, a plateau in statistical cure rates is observed within 10 years. Yet, for slower proliferating cancers, such as thyroid and early breast cancers, reaching this plateau might take even decades. We argue that a timeframe shorter than 10 years might overlook significant longterm impacts of cancer within the family. The course of the illness, late treatment effects, and possible recurrence can have enduring emotional impacts on family members (Mellon et al. 2007; Guan et al. 2021). Additionally, the effects of health shocks on families often persist beyond 4 or 5 years after the event.3Therefore, a 10-year post-cancer follow-up period is justified. This decision rules out the control group diagnosed with cancer in 2006 as a potential control group, leaving =11 and unaffected matched control group as potential candidates for constructing the counterfactual. We opted for FN DiD following recent 3For example, Fadlon and Nielsen (2019) demonstrate that a health shock in one family member affects the health behaviors of others for more than 4 years post-shock, and Vaalavuo et al. (2023) find that the adverse impact of child cancer on a mother’s earnings does not dissipate within 5 years post-shock. 123 A family affair? Long-term economic and mental health… Page 15 of 30 19 Table 1 continued Men (wife has cancer) Women (husband has cancer) Variable Control group Treatment group Adj. Difference P Control group Treatment group Adj. Difference P Psychiatric outpatient visits 0.008 0.009 0.001 0.254 0.013 0.014 0.002 0.127 Psychotropic medication 0.085 0.093 0.000 0.871 0.136 0.156 0.007 0.018 Antipsychotic medication 0.009 0.010 0.000 0.938 0.013 0.016 0.002 0.077 Anxiolytic medication 0.022 0.025 0.001 0.631 0.035 0.038 0.000 0.902 Hypnotic/sedative medication 0.027 0.032 0.000 0.948 0.043 0.056 0.005 0.005 Antidepressant medication 0.044 0.048 0.001 0.512 0.082 0.092 0.005 0.062 Opioid medication 0.020 0.021 −0.001 0.240 0.019 0.021 0.000 0.930 Charlson comorbidity index 0.110 0.114 −0.008 0.027 0.105 0.125 0.008 0.038 Spouse dies within 10 years 0.000 0.255 0.254 <0.001 0.000 0.437 0.438 <0.001 N 34484 23298 35849 20264 Notes: This table presents sample means for the treatment and the control groups, broken down separately for men and women, for the year preceding the index diagnosis. Columns 3 and 7 report the age-adjusted mean difference in the background characteristics between the treatment and control groups, and columns 4 and 8presentthe corresponding p-values. The sample comprises adults aged 28–64 whose spouses were diagnosed with cancer in Finland during the periods 2000 to 2008 (treatment group) and 2011 to 2019 (control group) 123 19 Page 16 of 30 P. Böckerman et al. individual, i.e., the breadwinner status. Following Bünnings et al. (2021); Foster and Stratton (2021), we define a breadwinner as a spouse who out-earns the other. We constructed a binary variable that assumes a value of 0 if the individual’s income contribution share was below 50% (indicating the secondary earner) 1-year prior to the index diagnosis and 1 if the share was above 50% (indicating the main breadwinner). Table 1describes the study sample, comparing the treatment and control groups before the index diagnosis. The index diagnosis denotes the year of cancer diagnosis for treated individuals and a placebo diagnosis year for the control group, occurring 11 years prior to the actual diagnosis year of the control group. Men (husbands of the cancer patient) were predominantly in the breadwinner category (72%), while women (wives of the cancer patient) were less represented (31%). Notable differences between the groups include employment status, retirement probability, and health, as measured by the Charlson comorbidity index (Charlson et al. 1987) using hospital data from 1996 until the year preceding the index diagnosis. These disparities are largely due to the different age distributions between the groups. Adjusting for a birth year reduces the average differences in the background characteristics, though some small differences remain, such as in earnings, disposable income, the likelihood of tertiary education, and health. These variations justify the use of difference-in-differences estimation with individual fixed effects. Cancer-related mortality reduced the household size by one in 34% of families during the follow-up period within the treatment group. Notably, men who were diagnosed with cancer had a considerably higher likelihood of death (44%) compared to women (26%) during the 10-year post-cancer follow-up period. 4 Results 4.1 Main effects of spousal cancer We start our empirical analysis by presenting the overall results based on Eq. 1. The estimates from this specification, accompanied by the corresponding 95% confidence intervals are depicted graphically in Fig. 2, while the parameter coefficients are reported in Appendix Tables A1–A4. The figures plot the change in the outcomes of interest relative to the year before spousal cancer diagnosis in the treatment group. Importantly, the figures do not exhibit clear pre-trends, thereby supporting the key identification assumption underlying our empirical specification. For women, we find evidence that they reduce their employment in the shortterm but increase employment in the long-term after their spouse’s cancer diagnosis (Panel A of Fig. 2). The average difference-in-differences estimate for women is zero (Appendix Table A2). Among male spouses, the effect was indistinguishable from zero throughout the follow-up. Overall, the labor market impacts are close to zero, which in contrast to the results in Jeon and Pohl (2017) for Canada according to which spouses of the cancer patients decrease substantially their labor supply (about 2 to 3 pp.) at the extensive margin leading to a lower level of earnings. Moreover, the changes 123 A family affair? Long-term economic and mental health… Page 17 of 30 19 A. Employment -2 -1 0 1 2 Effect on employment (pp.) -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 Years since diagnosis Men (wife has cancer) Women (husband has cancer) B. Retirement -3 -2 -1 0 1 Effect on retirement (pp.) -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 Years since diagnosis Men (wife has cancer) Women (husband has cancer) C. Earnings -3 -2 -1 0 1 2 Effect on earnings (%) -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 Years since diagnosis Men (wife has cancer) Women (husband has cancer) D. HH disposable income -15 -10 -5 0 Effect on HH disp. income (%) -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 Years since diagnosis Men (wife has cancer) Women (husband has cancer) E. Psychotropic drug use -2 0 2 4 Effect on psychotropic medication (pp.) -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 Years since diagnosis Men (wife has cancer) Women (husband has cancer) F. Psychiatric outpatient visits -.5 0 .5 1 Effect on psychiatric outpatient visit (pp.) -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 Years since diagnosis Men (wife has cancer) Women (husband has cancer) Fig. 2 The effect of spousal cancer on labor supply, income, psychotropic drug use, and psychiatric outpatient visits. The point estimates, with the shaded areas indicating the 95% confidence intervals, represent the differences in outcomes compared to the control group relative to the year preceding the index cancer diagnosis. The control group is composed of individuals diagnosed with cancer 11 years later relative to the treatment group. The vertical line at time r=−0.5 depicts the moment of the cancer diagnosis. Standard errors are clustered at the individual level. The corresponding event study estimates are detailed in Tables A1–A4 123 19 Page 18 of 30 P. Böckerman et al. in employment for women were reflected in the probability of retiring from the labor market (Panel B). Consistent with the employment effects, the effect on the probability of retiring was larger for women compared to men in the long-term. While this result contrasts with that of regarding Nordic countries in Jolly and Theodoropoulos (2023), it is in accordance with a negative effect on (early) pension benefits among widows found by Fadlon and Nielsen (2021). As the social insurance structure is very similar between Denmark and Finland, and data and empirical strategy are almost identical, the concordant results with Fadlon and Nielsen (2021) are unsurprising. However, when it comes to annual earnings (Panel C), the magnitude of the effect was negligible for women. Additionally, for male spouses, their earnings decreased after their spouse fell ill. This finding is consistent with the family effect, although the effect was modest with less than a 2% reduction in earnings. The estimates for the household’s disposable income (Panel D) (as well as for household’s total income and household disposable income adjusted by using OECD modified equivalence scale shown in Appendix Tables A1 and A2) reveal a meaningful decrease, from 10 to 15% in the medium to long run among female spouses and approximately 5% among the male spouses. This result reveals that a female spouse’s increase in earnings does not compensate for the loss of the sick spouse’s income within the household. Our results highlight that financial consequences following a spouse’s death tend to be harsher for women, which might explain the finding that cancer leads to a larger decrease in owner-occupancy in housing among women but not among men (Appendix Tables A1 and A2). Women may need to liquidate their assets to support household income and maintain material well-being. Importantly, in addition to consequences in the labor market, cancer also affects the spouse’s mental well-being. According to Panel E of Fig. 2, the probability of using psychotropic medication initially increased by approximately 4 percentage points (pp.) for women and about 2 pp. for men, and the impacts stabilized at around 1.5–2 pp. for both sexes. Relative to the baseline probability of psychotropic medication use, the relative increase in psychotropic medication was 13.0% for men and 14.1% for women on average during the full follow-up period. This result is in line with the evidence from Sweden and Denmark, suggesting a 13% risk increase in psychiatric disorders following a spouse’s cancer (Hu et al. 2023). Appendix Tables A3 and A4 also report the corresponding estimates regarding sub-categories of psychotropic medication such as antipsychotic, anxiolytic, antidepressant, and sleep (hypnotics/sedatives) medicine. Overall, the dynamics of the effects are largely similar to psychotropics in general. These tables also report increases in the probability of visiting psychiatrists in specialized public health care. This outcome indicates more severe psychiatric symptoms. The increase is larger for women (DD estimate 0.3 pp. vs. baseline 1.1 pp.) than for men (DD-est 0.2 pp. vs. baseline 0.8 pp.). Hereafter we report psychiatric impacts only in (any) psychotropic medication. A potential concern regarding the validity of the estimates is the endogenous nature of cancer. The event study specification ensures that the comparisons are conducted for individuals of the same sex, age, and cancer type of the spouse, and education level but with the timing difference of the spouse’s cancer diagnosis of 11 years. The main concern therefore is related to the timing of the diagnosis. The timing difference can 123 A family affair? Long-term economic and mental health… Page 19 of 30 19 potentially reveal differences between household living conditions and their health behaviors. Hence, as a robustness check, we re-estimated the effects of spousal cancer using only a subset of cancer diagnoses that are less related to health behaviors. This subset of cancers includes (ICD-10 categoryin parentheses): Gallbladder cancer (C23), Breast cancer (C50), and Ovarian cancer (C56), Prostate cancer (C61), Testicular cancer (C62), Thyroid cancer (C73), Myeloma (C90), Non-Hodgkin lymphoma (C8285,C96), Leukaemia (C91-C95), and Brain and other central nervous system cancers (C70-72). The selection of cancers was based on the British (Brown et al. 2018) and Australian (Wilson et al. 2018) estimates of the fraction of cancers that are preventable within each cancer diagnosis category. In our analysis, we adopted a conservative approach and included only those types of cancer estimated to be preventable by up to 30%. Encouragingly, the results based on this subsample (Appendix Tables A5 and A6) are by and large very similar to the baseline results presented in Fig. 2.Asan additional robustness check, we also restricted the sample to couples who had lived together for the entire 5-year period and found that results were quantitatively very similar (Appendix Tables A7 and A8). Moreover, we examined whether the use of an event study approach affects the interpretation of the main results. Appendix Fig. A4 shows the results of this exercise for our preferred choice, FN DiD with a control group affected by cancer 6 years later, using a stacked event study design and matching with unaffected households. We find that the estimates are qualitatively the same across all four approaches and, for the most part, also quantitatively very similar. As previously stated, the cancer survival rates differed notably by gender. 44% of men and 26% of women died during the 10-year follow-up (Table 1). This could lead to potential differences in the spousal labor supply responses between genders. For this reason, we next proceed to estimate the effects separately by the survival of the cancer patient. 4.2 Heterogeneity by breadwinner status in non-fatal cancers We examine whether the pre-cancer relative income status within the household influences the impacts of spousal cancer in non-fatal cancers. We separately estimated the breadwinner heterogeneity effects for female and male spouses using Eq. 2on earnings, employment, and psychotropic drug use in the short-, mediumand longterm. These results are presented in Table 2. The impact estimates for employment, psychotropic drug use, and marital status are presented in percentage points and as a percentage relative to the baseline values. The pre-cancer mean of the outcome within the breadwinner status is reported in the rightmost column. We find that there is a reduction in earnings for secondary earners both for men and women in non-fatal cancers. In secondary-earner women, earnings decrease by 3 % throughout the follow-up period. In contrast, secondary-earner men experience an initial decrease of 2%, followed by income deficits of 4% and 3% in the mediumand long-term, respectively. The point estimates are negative also for breadwinners but they are not statistically significant. The difference in earnings responses by breadwinner status is statistically significant for women in the short-term and suggestively different 123 19 Page 20 of 30 P. Böckerman et al. Table 2 Effects of spousal cancer by breadwinner status in non-fatal cancers Breadwinner status Short-term Medium-term Long-term Control group mean Est. % Est. % Est. % A. Women (husband has cancer) Earnings (%) Breadwinner −0.6[0.8]−0.7[1.1]−2.7[1.4]27549.1 Secondary earner −2.8[0.7]*−3[0.9]−3.3[1.1]16880.1 Employment (pp.) Breadwinner 0 [0.5]0−0.2[0.6]−0.3−0.4[0.7]−0.6 72.3 Secondary earner −1.1[0.4]†−1.8−0.3[0.4]−0.6−0.2[0.5]−0.459.2 Psychotropic Breadwinner 0.7 [0.5]4.7 −0.1[0.5]−0.40.4[0.6]2.7 15.4 drug use (pp.) Secondary earner 1.5 [0.3]8.7 1.3 [0.3]*7.21.1[0.4]6.4 17.4 Married with Breadwinner 0.2 [0.3]0.3 0 [0.4]0−0.2[0.5]−0.3 84.7 cancer patient (pp.) Secondary earner 0.6 [0.2]0.6 0.9 [0.3]†1 0.9[0.3]†1.1 89 B. Men (wife has cancer) Earnings (%) Breadwinner 0.1 [0.4]−0.7[0.6]−1.5[0.8]36195.2 Secondary earner −1.8[1.3]−3.6[1.5]†−3.3[1.7]15184 Employment (pp.) Breadwinner −0.2[0.3]−0.3−0.3[0.3]−0.5−0.1[0.4]−0.171 Secondary earner −0.1[0.5]−0.10[0.6]0−0.2[0.7]−0.4 52.2 Psychotropic Breadwinner 0.6 [0.2]5.6 0.4 [0.2]3.5 0.6 [0.3]6.2 10.2 drug use (pp.) Secondary earner 0.6 [0.4]4.4 0.7 [0.4]5.3 0.8 [0.5]5.9 13.2 Married with Breadwinner 0 [0.2]0−0.2[0.3]−0.20[0.3]087.1 cancer patient (pp.) Secondary earner 0.3[0.3]0.4 0.6[0.4]0.7 0.1[0.5]0.1 83.2 Notes: Short-, medium-, and long-term impacts of spousal cancer by different breadwinner status in non-fatal cancers. Standard errors (clustered at individual level) are reported beside the point estimates in parentheses. Short-term refers to DD estimates using post-event periods 0–2, medium-term to periods 3–5, and long-term to periods 6–10. Symbols †, * and ** refer to statistical significance 10%, 5% and 1% of the point estimates relative to the reference group (Breadwinner). All estimates are based on the triple-difference models presented in Eq. 2 123 A family affair? Long-term economic and mental health… Page 21 of 30 19 for men in the medium-term. Interestingly, the responses are considerably more subtle in terms of employment, being negative only for secondary-earner women in the shortterm. This suggests that in non-fatal cancers, the spouses of the patients adjust their labor supply more on the intensive margin than on the extensive margin, in line with the evidence presented by Fadlon and Nielsen (2021). The purchases of prescribed psychotropic medications increase for both men and women. The effects are most pronounced in secondary earners, with a 1.5 percentage point (pp) increase in the short-term, and 1.3 pp. and 1.1 pp. increases in the medium and long-term, respectively. Heterogeneity in relative income is statistically significant only for women in the medium-term. A potential factor affecting economic and mental health responses is union stability. In the final panels of Table 2, we report the impact of cancer on being married with the cancer patient. Approximately 87% of the couples were married at the baseline. We find that cancer has a positive effect on marriage for the secondary earner (cancer patient is the breadwinner) and no effect on the breadwinner (cancer patient is the secondary earner). This result is in line with prior research by Ehlert (2021) that found that health shock increases the probability of marriage among cohabiting couples more the higher are expected survivor pension for the widow. However, no such effects are found for men. 4.3 Heterogeneity by breadwinner status in fatal cancers Finally, we investigate whether the impact of a cancer diagnosis resulting in death within the 10-year follow-up period differs by breadwinner status. In this analysis, we exclude marital outcomes, as death terminates marriage in the treatment group unless it had already dissolved beforehand. The analysis is based on a sample from which we exclude households where the cancer patient did not die during the follow-up period. These results are reported in Table 3. We find that fatal spousal cancer initially decreases earnings by 3% and 2% for secondary and breadwinner women, respectively. In the long-term, there is a considerable contrast in the response in terms of relative income. Earnings of breadwinner women remained unchanged, whereas those of the secondary earners exhibited a 6% increase. A similar pattern is observed at the extensive margin; breadwinners showed no change in employment, while secondary earners experienced a significant 4 pp. increase in the long-term. The labor market responses were statistically significantly different for the secondary earner and breadwinner women only in the long-term. However, the increase in the probability of psychotropic medication purchases was consistently higher (1.4 pp. to 1.9 pp. higher) for the secondary earners compared to breadwinner women. The breadwinner heterogeneity was markedly different for men. For men, we found no statistically significant differences by the breadwinner status for the most part, but the coefficients point towards statistically significant responses in the long-term with breadwinner men decreasing and secondary earners increasing their labor supply at the intensive margin. Additionally, breadwinner men showed higher increases in 123 19 Page 22 of 30 P. Böckerman et al. Table 3 Effects of spousal cancer by breadwinner status in fatal cancers Breadwinner status Short-term Medium-term Long-term Control group mean Est. % Est. % Est. % A. Women (husband has cancer) Earnings (%) Breadwinner −2[0.8]−0.1[1.1]−0.2[1.3]27398.9 Secondary earner −2.7[0.8]1.8[1]6.2[1.2]** 16709.7 Employment (pp.) Breadwinner −0.3[0.6]−0.50.5[0.7]0.7 0 [0.7]072 Secondary earner −0.3[0.4]−0.61.7[0.5]2.9 4.3[0.6]** 7.3 58.8 Psychotropic Breadwinner 6.4[0.6]41.7 4.2[0.6]27.5 2.7[0.6]17.4 15.3 drug use (pp.) Secondary earner 7.8[0.4]* 44.8 6.1[0.4]** 35.4 4.3[0.5]* 24.5 17.3 B. Men (wife has cancer) Earnings (%) Breadwinner −1.9[0.7]−3.1[0.9]−2.3[1.1]36131.4 Secondary earner −2.5[2]−2.1[2.2]2.2[2.5]† 15116.9 Employment (pp.) Breadwinner −0.7[0.5]−1−0.6[0.5]−0.8 0 [0.6] 0 70.9 Secondary earner 0.7[0.8]1.3 0.4[0.9]0.7 1.1[1.1]2.1 52.1 Psychotropic Breadwinner 4.3[0.4]42.2 4 [0.4]39.3 3.8[0.5]37.2 10.2 drug use (pp.) Secondary earner 2.9[0.6]* 21.7 4.6[0.7]34.7 3.3[0.8]24.9 13.2 Notes: This table presents the short-, medium-, and long-term impacts of spousal cancer by different breadwinner statuses in fatal cancers. Standard errors, clustered at individual level, are presented alongside the point estimates in parentheses. Short-term refers to DD estimates for post-event periods 0–2, the medium-term to periods 3–5, and the long-term to periods 6–10. Symbols †, * and ** denote statistical significance 10%, 5%, and 1% levels, respectively, for the point estimates relative to the reference group (Breadwinner). All estimates are derived from the triple-difference models presented in Eq. 2 123 A family affair? Long-term economic and mental health… Page 23 of 30 19 psychotropic medication purchases in the short-term compared to secondary earners, but not in the mediumand long-term. Overall, we find that death is the primary driver behind the increased psychotropic drug use among spouses of cancer patients. When the cancer patient survives, spouses are relatively unaffected, but fatal cancer leads to notable increases in psychotropic drug purchases. While the estimates are not directly comparable, our results align with those of Angelini and Costa-Font (2023), which suggest that fatal cancer leads to a substantial increase in self-reported depression symptoms, while non-fatal cancers exhibit more subtle changes in psychological symptoms among spouses. 4.4 Reconciling the evidence across the relative income distribution Overall, the results from the previous section suggest that the role of relative earnings is stronger for women than for men in responses to spousal cancer. However, a binary indicator for breadwinner status, determined strictly by a 0.5 cut-off does not capture the subtle responses along the relative income spectrum. To better understand the role of relative earnings in the responses to spousal cancer, we divided the sample into five equal-sized groups by the pre-shock earnings income share of the individual. For illustrative purposes, we focus solely on long-term responses, i.e., the impacts of spousal cancer on outcomes measured 6 to 10 years after the cancer diagnosis. Essentially, we conducted difference-in-differences analyses to estimate the long-term impacts separately for each relative income quintile and by sex. Figure 3shows the point estimates along with their 95% confidence intervals. On the x-axis, the income quintile shares represent the spouse’s share of the total household income 1 year prior to the diagnosis. The mean values of income shares within each quintile are reported in the parentheses. We find that the lower a spouse’s relative income share within the household, the larger the increase in their earnings. The relationship is nearly linear in terms of earnings for both men and women (as shown in Panels A and B). This suggests that greater income losses due to losing a spouse correlate with a larger increase in labor supply at the intensive margin, relative to the counterfactual trajectory. This result aligns with the evidence from Fadlon and Nielsen (2021), which shows that the amount of a spouse’s income lost due to fatal cardiovascular events is positively linked to labor supply responses among surviving spouses. Notably, our analysis also suggests that this pattern exists in both fatal and non-fatal cancers. The heterogeneity in the responses at the extensive margin of labor supply is less evident (Panels C and D). We find that a spouse’s baseline income share is negatively correlated with the added worker effect but only for women in fatal cancers. For both men and for women with non-fatal cancers, relative income does not appear to influence employment responses. Collectively, these results suggest that men are more likely to adjust their labor supply at the intensive margin, whereas for women, the extensive margin accounts for a larger share of the changes in earnings following a spouse’s cancer diagnosis. 123 19 Page 24 of 30 P. Böckerman et al. A. Women’s earnings -12 -6 0 6 12 18 24 Effect on earnings (%) 1 (0.225) 2 (0.400) 3 (0.493) 4 (0.587) 5 (0.764) Baseline (own) income share quintile Non-fatal cancer Fatal cancer B. Men’s earnings -12 -6 0 6 12 18 24 Effect on earnings (%) 1 (0.225) 2 (0.400) 3 (0.493) 4 (0.587) 5 (0.764) Baseline (own) income share quintile Non-fatal cancer Fatal cancer C. Women’s employment -6 -4 -2 0 2 4 6 Effect on employment (pp.) 1 (0.225) 2 (0.400) 3 (0.493) 4 (0.587) 5 (0.764) Baseline (own) income share quintile Non-fatal cancer Fatal cancer D. Men’s employment -6 -4 -2 0 2 4 6 Effect on employment (pp.) 1 (0.225) 2 (0.400) 3 (0.493) 4 (0.587) 5 (0.764) Baseline (own) income share quintile Non-fatal cancer Fatal cancer E. Women’s psychotropic medication -4 -2 0 2 4 6 8 Effect on psychotropic medication (pp.) 1 (0.225) 2 (0.400) 3 (0.493) 4 (0.587) 5 (0.764) Baseline (own) income share quintile Non-fatal cancer Fatal cancer F. Men’s psychotropic medication -4 -2 0 2 4 6 8 Effect on psychotropic medication (pp.) 1 (0.225) 2 (0.400) 3 (0.493) 4 (0.587) 5 (0.764) Baseline (own) income share quintile Non-fatal cancer Fatal cancer G. Women’s marriage -4 -2 0 2 4 Effect on staying married (pp.) 1 (0.225) 2 (0.400) 3 (0.493) 4 (0.587) 5 (0.764) Baseline (own) income share quintile Non-fatal cancer H. Men’s marriage -4 -2 0 2 4 Effect on staying married (pp.) 1 (0.225) 2 (0.400) 3 (0.493) 4 (0.587) 5 (0.764) Baseline (own) income share quintile Non-fatal cancer Fig. 3 Spouses’ long-term response estimates along with 95% confidence intervals (years 6–10 after vs. 5 years preceding the cancer diagnosis) to fatal cancer, categorized by their pre-shock income share quintile within the household’s total earnings. We divided the sample into five equal-sized groups based on the spouses’ pre-cancer share of the total household income. Subsequently, we plotted the average outcome response against the pre-cancer mean income share for each quintile 123