Men who care too much do not work: a systematic review and meta-analysis of the effect of unpaid caregiving by men on their labour market participation

Objective: We conducted a systematic review and meta-analysis to assess the effect of unpaid caregiving by men on their labour force participation (LFP). Methods: English-language journal articles, dissertations and working/conference papers from the MEDLINE, Embase, AgeLine, EconLit, EconPapers and the International Bibliography of the Social Science databases were searched from January 2007 to February 2021. We pooled the marginal effect of caregiving on men’s LFP in a random-effects model where LFP was dened as being gainfully employed while non-LFP comprised all other circumstances. Non-parametric methods were used to estimate the probability that caregiving was negatively associated with LFP. Meta-regressions were conducted with countries stratied by Gross National Income (GNI) per capita as a moderator variable. Secondary analyses were performed on employed men to assess the potential caregiving effect on temporary work exit and permanent retirement. Results: Thirty-one studies (904,694 men) were included in the review with 28 of these studies (892,805 men) used in the meta-analysis. The random-effects model found men who were weekly, daily or primary caregivers to experience a 4.4%, 8.0% and 10.0% reduction in their LFP, when compared to their respective counterparts. We estimated weekly, daily or primary caregivers to have an 87.3% (95% CI: 70%-96%), 87.5% (95% CI: 84%-96%) and 80.8% (95% CI: 67%-100%) chance of a lower LFP. An hourly increase in caregiving was unrelated to LFP (p-value=0.106). Among employed men (24,319 men), caregiving was found to be unrelated with retirement (p-value=0.156) and no meta-analysis was conducted on temporary work exit. No difference was found across countries stratied by GNI per capita. Conclusions: These results urge policy decisionmakers to provide more exible work arrangements and other protective measures to support male employees who are weekly, daily or primary caregivers. Future meta-analyses should explore the effect of caregiving on other dimensions of labour supply (e.g., hours of work) by men. The scarcity of studies conducted in low-/middle-income countries might be addressed by creating population-based cohorts following the protocol of existing studies such as the American HRS (Health and Retirement Study).


Introduction Background
Caregiving has traditionally been regarded as "women's work", although the number of male caregivers is on the rise. 1 Just before the coronavirus pandemic, 40% of unpaid caregivers aged 50+ across OECD countries were men. 2 During 2019, 21% of American adults indicated that they provided some unpaid care, with 39% of these caregivers being men. 3 A similar distribution of caregivers by sex has been noted in Australia and Europe, with a few countries (such as Sweden) reporting equivalent, or even slightly more, male than female unpaid caregivers. 4,5 Among outcomes that may be adversely impacted by caregiving, labour force participation (LFP) is particularly important, as the working-age population is already shrinking at an alarming rate due to population ageing. 6 Since men comprise more than half of the working population and are experiencing a faster decline in their rates of employment than women, 7 it is crucial to assess if, and to what extent, caregiving might hinder the LFP among the general male population; and furthermore, how it might lead to temporary withdrawal from the labour market or permanent work exit (i.e., retirement) among men that are employed.
While most examinations of the relationship between unpaid caregiving and work are performed in high-income countries (HICs), especially Western countries, there are opportunities to explore the unique circumstances facing this trade-off for men in other regions of the world. Indeed, the unique social and economic structures of low-/middle-income countries (LMICs) might imply that men face inherently different pressures when balancing work and caregiving. First, the demand for home-based informal care is very high in LMICs due to the limited supply of public formal care, especially for those services that substitute for informal caregiving. 8,9 Second, owning to improved education, maternal health, and shifts in social norms and family roles, the majority of LMICs have seen a faster increase in the share of women in the labour force compared to HICs. 10 Indeed, one of the unintended consequence of a surging LFP among women is a potentially heavier burden of caregiving placed on men. Third, many LMICs have highly unbalanced gender ratios in the society, a result of gender selection through prenatal sex determination and selective abortions. [11][12][13] As men signi cantly outnumber women, wifeless single men are potentially tasked with heavy eldercare duties. And nally, LMICs tend to have less government or employer supported family-friendly policies, while having laws that mandate unpaid caregiving for the elderly (observed in countries such as China 14 ). These observations warrant investigations into the association between care provision and LFP in men from LMICs to understand how they seek out the economic trade-off between paid work and unpaid caregiving.
Building upon the results of previous reviews on the burden of men who were caregivers 15,16 and reviews that have speci cally dealt with the association between caregiving and labour supply [17][18][19][20] , we conducted this systematic review and meta-analysis to assess the following three questions: Q1. Does unpaid caregiving impact men's LFP?
Q2. Among employed men, does care provision in uence the risk of temporarily withdrawing from the labour market or exiting work permanently (retirement)?
Q3. Is there a differential effect of caregiving on men's LFP between HICs and LMICs?
Answers to these questions can potentially inform strategies that enhance the work prospects of men who perform caregiving duties and may help identify tailored policy to assist men in their balancing of work and caregiving, in general, and for men in LMICs, in particular.

Theoretical framework
The economic theories of specialization, bargaining and the division of labour are commonly used to explain the higher LFP of men over women. [21][22][23] Notably, Jacob Mincer modelled women's time allocation across leisure, home production, and labour work, while restricting the choice for men to be between just work and leisure. 24 Based on the concept of the reservation wage, women would compare their market wage to the value of their time allocated to home production or leisure, [25][26][27] whereas the comparison faced by men was between their market wage and the value of their leisure time. Regardless of gender, the standard labour-leisure choice model is often invoked to theorize individual decisions between unpaid caregiving and paid work. This model states that when confronted with caregiving duties, individuals maximize their overall well-being by allocating time spent in ways that equalize the return on time allocated to alternative activities, i.e., the concept of equimarginal returns is invoked. [28][29][30] This concept when applied to the decision to work (or not) suggests that when holding all else constant, an increase in caregiving intensity (that is usually operationalized as the hours of caregiving) results in an availability effect that tends to lower individuals' LFP, resulting in temporary work exit or potentially permanent retirement. The caregiving-LFP relationship also depends on individual preference orderings between labour and leisure, social context, the life cycle and other factors. 31,32 The association can also take on more complex mathematical forms that involve both kinks and/or discontinuities if a caregiving threshold(s) is deemed plausible. 33

Study eligibility
This study is registered on the Open Science Framework and follows the Meta-analyses of Observational Studies in Epidemiology (MOOSE) guideline. 34 We searched for empirical studies that were published between January 1, 2007 and February 15, 2021 to avoid those that had been repeatedly reviewed. [17][18][19][20] Studies were included when they examined the relationship between unpaid caregiving by men and their LFP (i.e., whether they were gainfully employed) in a regression analysis. We also included studies of strictly employed men where they assessed the association between caregiving and work exit (i.e., temporarily quitted working) or retirement (i.e., permanently ceased working). Studies on the intention to participate in the labour market or only assessing the reciprocal association (i.e., how LFP in uenced men's care provision) were excluded. Studies that only examined living arrangements (such as living with grandchildren) or the presence of a health condition (such as having a spouse with cancer) without explicitly measuring men's caregiving activities were excluded as well. Further exclusions were performed for simulation studies, qualitative studies and reviews, although their reference lists were inspected.

Literature search
One reviewer (RF) conducted literature search on February 15, 2021 using a combination of subject headers and keywords on MEDLINE, Embase, AgeLine, EconLit, EconPapers and the International Bibliography of the Social Science databases. The search strategies were designed in consultation with a librarian. English-language journal articles, working/conference papers and dissertations were included. The same reviewer applied the snowball method to track additional papers outside the protocol-driven database search strategies. 35 These citations were uploaded to the Covidence platform where two reviewers (RF and JG) independently screened the abstract, title and the full text to determine their eligibility. Disagreements were resolved through discussion.

Measuring caregiving
Because studies varied extensively in their measurement of caregiving, we categorized four types of caregiving-including any caregiving, daily caregiving, primary caregiving, and intensive caregiving-using de nitions drawn from existing studies. Any caregiving refers to the provision of at least 100 hours of care over the past 2 years (or approximately, at least an hour of caregiving per week).
This de nition was drawn from the American HRS (Health and Retirement Study) that has been used in the European SHARE (Survey of Health, Ageing and Retirement in Europe) as well. 36,37 Daily caregivers were those who either reported spending 10 hours or more per week (or about 1,000 hours or more over 2 years) on caregiving or reported the performance of caregiving on a daily basis. Primary (or main) caregivers were de ned as those who self-identi ed as such. No consensus has been reached in the literature regarding the de nition of intensive caregivers. 19 In order to maintain consistency within this review and with the most recent systematic review on this topic, 19 we used 15 hours or more of caregiving per week to de ne intensive caregivers and by doing so distinguished them from daily caregivers who were de ned by a threshold of 10 hours or more of care per week. Hours of caregiving was considered as another measurement of caregiving. The care tasks involved, the identity of the care recipients, the status of being a new caregiver, and the coresidential status of the caregiver with the recipient were explored narratively.

Data extraction
Studies were included in the meta-analysis if they reported the marginal effect and the associated standard error of unpaid caregiving on men's LFP. The marginal effect corresponds to the coe cient estimate of caregiving in a linear probability model estimated using the OLS (ordinary least-squares) method or it is the average ceteris paribus probability effect of caregiving on being in the labour market estimated from a probit or logistic regression. 38 For studies that did not report a marginal effect, we reached out to the authors to request the availability of such an estimate and received two replies from seven requests. For studies that used instrumental variables to overcome the potential endogeneity of caregiving, we extracted the corresponding marginal effect estimates unless the study failed to reject exogeneity.

Random-effects meta-analysis and meta-regression
For the caregiving categories and for hours of caregiving, we used separate random-effects models to pool the marginal effect estimates to yield an average caregiving effect on men's LFP. The random-effects models were estimated using the restricted maximum likelihood methods with the Hartung-Knapp standard error adjustment to mitigate potentially high heterogeneity. 39 For each set of analysis, publication bias was assessed in a funnel plot with the Egger's test using the 5% threshold for statistical signi cance. 40 Non-parametric methods were applied to estimate the probability of men experiencing a reduction in LFP due to caregiving. 41,42 Metaregressions were conducted to explore the presence of a differential caregiving effect on LFP between HICs and LMICs. This was performed by dichotomizing the place of data origin to be either a HIC or a LMIC using the World Bank's US$12,536 threshold for percapita Gross National Income. 43 Secondary analysis and risk of bias assessment Two secondary analyses were carried out for the outcomes of temporary work exit and retirement, respectively, using studies of strictly employed men. A modi ed Newcastle-Ottawa Scale for cohort studies was applied to assess the quality of evidence. 44 Speci cally, we replaced the "study has controlled for the most important factor" item with whether studies have utilized any techniques to mitigate the potential endogeneity of caregiving. We also removed the item that required studies to not include any individuals who presented with the outcome at the start of the study. Thus, studies could score 0-8 in total, where we deemed scores of 7-8, 5-6 and below 5 to indicate low, medium or high risk of bias.

Study selection
The literature review yielded 2,189 citations, of which 1,661 were unique ( Figure 1). Twelve additional studies were found using snowballing, leading to 1,673 studies for title and abstract screening. The full texts of 182 studies were subsequently assessed. We excluded 152 studies for the following reasons: duplicates (n=1) or published in non-English (n=1); did not assess LFP (n=50) or only assessed anticipated LFP (n=1); qualitative studies (n=16), reviews (n=7), simulation studies (n=1), or editorials or governmental reports (n=7); involved only women (n=14) or did not report results for men (n=9); did not assess the caregiving-LFP relationship (n=20), only assessed the reciprocal relationship (n=12), or were unrelated to caregiving (n=12). These procedures yielded 31 studies for our review of which 28 were used in the meta-analysis.

Characteristics of the included studies
The 31 studies drew data from 904,694 (non-unique) men from several regionally based cohorts (Table 1). European-based studies comprised half of all studies (n=14), followed by seven studies using data from North America, another seven conducted in Asia, and three in Oceania (Australia). Only three studies were based in a LMIC, including two from China 45,46 and one from Indonesia 47 . Sample sizes of men ranged from 265 to 768,125. While most (n=20) studies focused on men aged 45+, ve included men in their 20s and three even included men below 20 years of age. Fourteen studies focused on speci c care recipients, such as parents/parents-in-law (n=10) or only grandchildren (n=1) or both (n=3), while the remainder kept a wide range of care recipients.
The marginal effect of caregiving on men's LFP Twenty-three studies (876,880 men) estimated the effect of caregiving on LFP (Table 2). Fourteen studies used instrumental variables, wherein six con rmed caregiving to be endogenous. By excluding the three studies that did not report a marginal effect, 48-50 results from the remaining 20 studies (864,991 men) were entered into the meta-analysis. Random-effects models were estimated for any, daily, primary, and hours of caregiving, respectively, while a narrative synthesis of results on intensive caregiving is presented herein due to the scarcity of studies (n=2) that it precluded a meta-analysis. Meta-regressions were conducted for any caregiving and hours of caregiving, as no LMIC-based studies were identi ed for the other caregiving categories.
Two studies assessed the impact of intensive caregiving beyond 15 hours per week on LFP. A South Korean study found that compared to non-caregiving men, those providing 1-19 hours or at least 20 hours of caregiving per week experienced a 2.8-percentagepoint (95% CI: -0.162 to 0.106) and 9.3-percentage-point (95% CI: -0.276 to 0.090) insigni cant decrease in LFP, respectively. 54 Another large census-based UK study suggested that up to 20 hours of caregiving per week did not affect men's LFP; however, spending 50 or more hours per week on caregiving decreased their LFP by 24.9 percentage points (p-value<0.01). 60 Four studies accessed the effect of caregiving hours on men's LFP. 45,54,63,64 The random-effects model ( Figure 5) found each hourly increase in caregiving to be associated with an insigni cant 0.2-percentage-point decrease in LFP (95% CI: -0.0004 to 0.001, p-value=0.106). The level of heterogeneity was relatively low (I 2 =31.1%) and no publication bias was detected (p-value=0.813). There was a 37.2% chance (95% CI: 9.73% to 84.7%) of lower LFP with greater hours of caregiving. We also found via meta-regression that there was an insigni cant difference between men from HICs and LMICs (p-value=0.494).

Secondary analysis on employed men
Five studies (18,689 employed men) assessed the association between care provision and temporary work exit (Table 3). We did not perform any meta-analysis due to the use of odds ratios and/or hazard ratios in these studies. Among studies of new caregivers, the only signi cant association was found in the UK where caregiving men who were new co-residents with the care recipient experienced higher odds of work exit during the same year compared to men who were not caregivers (odds ratio=1.61, 95% CI: 1.02 to 2.20). 65 However, the hours of caregiving were unrelated to the risk of work exit among new caregiving men. 65,66 When the analysis was strati ed by distinct caregiving tasks, one UK study found those providing personal, basic or instrumental care were not at any higher risk of work exit than their non-caregiving counterparts. 67 Finally, two studies-based in Canada and the UK, respectively-found middle-aged male caregivers did not exit work at a higher rate than non-caregivers. 68,69 The effect of caregiving on decision to retire was assessed in ve studies involving 15,432 employed men (Table 4). A random-effects meta-analysis was performed on three of these studies and ruled out the association between any caregiving and retirement (p-value=0.156). 37,51,70 Two American studies found no effect of personal caregiving, chore caregiving, or spousal caregiving on men's retirement. 37,51 A German study also ruled out the effect of co-residence with a care recipient on retirement. 71 Three studies have evaluated the impact of hours of caregiving. 37,71,72 In Canada, providing up to 5 hours of care per week increased the relative risk that men would be fully retired over being fully employed (relative risk ratio=1.52, 95% CI: 1.05 to 1.99). However, the effect was absent among men providing more than 5 hours of care per week. 72 Appraisal of study bias The modi ed Newcastle-Ottawa Scale was used to critically appraise the quality of each study (Table 5). Studies scored generally high between 5-8 by using large nationally (or regionally) based panel data and reasonably measured caregiving and labour market outcomes. Points were deducted for studies relied solely on cross-sectional data and/or did not conduct any procedures to examine the potential endogeneity of caregiving.

Discussion
By systematically reviewing and meta-analyzing published results since 2007, our study provided answers to the three research questions: namely, (Q1) men who were weekly, daily or primary caregivers had lower LFP than their counterparts, while the LFP was unaffected by an hourly increase in caregiving; (Q2) caregiving did not impact men's decision to permanently retire, although no metaanalyses were conducted on temporary work exit due to the unavailability of marginal effect estimates from those studies; (Q3) while there were only 3 studies from LMICs included in our review, thereby lowering the statistical power of our analysis, we were unable to distinguish differences in caregiving-LFP relationship between men in HICs and LMICs. This is the rst meta-analysis to con rm a statistically signi cant reduction in LFP among men who were weekly, daily or primary caregivers. Both Lilly et al (2007) and Bauer & Sousa-Poza (2015) found little to no evidence of a lower LFP among caregivers, although they did not distinguish caregivers by gender nor did they classify caregivers by the intensity of care. 17,18 Moussa (2018) reviewed the literature on the LFP of unpaid caregivers of elders across OECD countries. 19 Although this review did concur with our ndings on a decreasing trend of LFP among more "intensive" caregivers, the focus was placed on women, as 30% (14 of 48) of the studies reviewed were conducted exclusively on women. Moreover, some ndings appeared to be left out from the discussion, particular the three studies that demonstrated reductions in LFP for men who cared for elders. 37,66,71 These observations imply that prior reviews may have inadequately captured the labour consequences of caregiving in men. As female caregivers dominate the empirical literature, the experiences of men are not well known, as re ected in these reviews. 15,16 Our results are consistent with the labour-leisure choice model that predicts that while small increases in caregiving time tend to cause reductions in leisure rather than work, larger incursions are often su cient to shift time away from work to result in non-participation in the labour market. According to our ndings, being tasked with at least weekly caregiving (that extends to daily or primary caregiving) could represent meaningful threshold as when men consider exiting from the labour market. Furthermore, when leisure is restored as a result of shifting time away from work, it is both desired in itself and may represent a resource that supports the provision of care, i.e., men might devote some leisure time to invest in the ability of caregiving. Hence, when the care demand is small, the required "training time" may also be small; however, when caregiving becomes more intensive and di cult, men may require substantial time out of their total leisure time to invest in caregiving ability, which may ultimately lead to labour force non-participation. 73 In our analysis, men who are weekly, daily or primary caregivers might allocate time from work to "learn" how to care. This is a plausible observation as previous ndings have shown men nd certain types of care work, such as personal care tasks including bathing, dressing and toileting to be extremely challenging. 15,16 Some men also nd seemingly easier tasks, such as cooking, to require extensive learning and practices as well. 16 We found that there was a scarcity of studies examining the caregiving-LFP association of men in LMICs. Thus, we were unable to con dently rule out the role of country type, even with the insigni cant results from meta-regressions. Most LMICs lack the data sources that are essential for this line of research, i.e., a large population-based cohort with detailed individual-level data on sociodemographic status, household structures, health, labour supply and unpaid caregiving behaviours. Another recently concluded review also deemed the volume of literature on caregivers from LMICs to be insu cient for a meta-analysis. 74 The ndings of this study offered insights for labour and social policies to enhance the work prospects of men who are unpaid caregivers. In order to retain these employees, employers might consider adopting family-friendly policies (e.g., exible work hours) and worker bene t (e.g., paternal leaves). 75 Educational programs may be designed to deliver coping strategies to support male employees responsible for caregiving. For example, men are shown to prefer problem-focused, task-oriented coping strategies rather than storysharing and emotional support; 1 hence, interventions such as support groups need to re ect the unique demands and approaches of male caregivers. 76 It is also important that these interventions do not strengthen caregiving as "women's work", as male caregivers may not seek help due to the constraints of holding traditional and more conservative masculine values. 16 Hence, messages that de ne caregiving within the realm of masculinity, such as those posited in Hanlon's framework of caring masculinities 77 , need to be popularized to ensure men perform care with a peace of mind. Finally, as men are generally more reluctant than women to seek out support in the course of caregiving, 1,16 employers and policy decisionmakers need to adopt an active approach to identify men who assume at least weekly or primary caregiver roles as they are likely subject to higher risks of work-life imbalance.
Research agencies in LMICs might consider establishing large population-based cohorts to allow for nationally representative analyses on caregiving and labour market outcomes. Ideally, the survey instruments need to be comparable with that of existing studies, such as the American HRS, the Chinese CHARLS or the European SHARE, to enable cross-country comparisons. Several years of follow-up on the cohort would allow for more sophisticated longitudinal analyses to assess the time trends of caregiving and labour supply, although repeated cross-sectional datasets would also be valuable. An alternative to creating new cohorts is to enhance the national census with items on caregiving and labour work, as seen in the Census for England and Wales. 60

Study limitations
Our study has several limitations. First, selection bias may arise from restricting studies to be those published in English. Hence, future researchers might wish to remove the language restriction in their search to conduct a more comprehensive synthesis. Second, the small number of studies has precluded us to perform a meta-analysis on some occasions or lead to insu cient statistical power on others. For example, we only identi ed two studies on intensive caregivers and thus did not conduct a meta-analysis. Furthermore, using the results of three studies, we demonstrated that LFP was negatively associated with being a primary caregiver. Despite the small number of studies, it is worth stressing that all three studies agreed on a signi cantly reduced LFP among primary caregiving men. Hence, we deem the results of the meta-analysis to be plausible as primary caregiving is likely to lead to a lower LFP for men.
Third, while there is a lack of consensus in the literature on the de nition of each type of caregiving, we used speci c de nitions of any, daily, primary and intensive caregivers in this work. [17][18][19] While our ndings are potentially sensitive to the de nitions employed, we adopted de nitions that are commonplace in the literature. It is clear that more standardization on such de nitions would facilitate the performance of robust meta-analyses. Fourth, we did not examine labour outcomes of men beyond LFP, such as hours of work or hourly wages. The labour decision of individuals is commonly modeled as a two-part process, wherein individuals rst decide whether or not to work, and conditional on the decision to work, they select the hours of work based on the principle of equimarginal returns, i.e., where the gain in the utility from enhanced income just offsets the marginal utility of leisure. 33 Hence, we believe our work to represent an important rst step in the evidence synthesis literature to understand how men balance caregiving and paid work. And nally, none of the studies in this review used data collected during the coronavirus pandemic. Therefore, we were unable to account for the potentially enlarged wealth gap between HICs and LMICs and other outcomes caused by the pandemic. 78 Hence, future researchers need to revisit this topic by using more recent data to re ect these impacts.

Conclusions
This systematic review and meta-analysis found men who were weekly, daily or primary caregivers to have a lower LFP rate than their counterparts. Although we did not nd country type to moderate the association between caregiving and men's LFP, these results may be due to the paucity of studies from LMICs. Our ndings support the design of tailored policies to retain employees who simultaneously assume weekly, daily or primary caregiver roles. In order to gain a more comprehensive understanding on the labour market consequences of caregiving, research agencies in LMICs are encouraged to consider the establishment of large populationbased cohort studies of caregiving and its impact on various labour market outcomes using instruments that are comparable to those of existing studies that address health and ageing.

Declarations
Competing Interests: The authors declare no competing interests. This work received no speci c grant from any funding agency in the public, commercial, or not-for-pro t sectors.  Carr (2016) † Individuals with at least one employment>6 month after schooling were included. ‡ The original paper reported the odds ratio of being employed relative to leaving employment. The reciprocal odds ratios were computed using the following formula: point estimate=1/OR with standard error=(1/OR) 2 SE(OR) using the delta method. ¶ Recipient is a relative that is not the spouse or the parent of the caregiver. § Recipient is anyone that is not the spouse or partner or parent or grandparent of the caregiver.
# Recipient is a non-relative of the caregiver.
Carmichael (2010), Carr (2016) and King (2013) assessed caregiving provided to anyone who was sick, elderly or disabled. Gomez-Leon (2017) studied parents or parents-in-law as the care recipients, while Proulx (2014) assessed care given to parents, grandparents, spouse, another relative or non-relative.