Abstract
This work explores how the parenthood wage penalty is partially explained by an increased within-couple gap in job uniqueness (that is, the within-establishment substitutability of workers). Uniqueness is rewarded with higher wages, but it requires worker presenteeism (that is, the lack of unpredictable work absences), which entails a higher cost of childbearing. Using a within-couple event study approach, we show that after the arrival of the first child, women take more days of absence than men, and their likelihood of holding jobs with low substitutability decreases. We find that 15 years after childbearing, the male–female gender gap in holding a (higher-paying) unique job increases by six percentage points. The results suggest that structural changes towards greater work flexibility, making it less costly for workers in unique jobs to not be present, can help to reduce the parenthood wage penalty.
I. Introduction
The way in which jobs and firms are structured is an important challenge for achieving gender equality in the labor market (Goldin 2014). Despite a great deal of convergence in the gender wage and employment gaps over the last century, important long-run economic gaps persist between similarly trained and educated men and women. Recent research highlights the importance of parenthood on the emergence and persistence of gender inequality in the labor market (Ejrnæs and Kunze 2013; Angelov, Johansson, and Lindahl 2016; Adda, Dustmann, and Stevens 2017; Kleven, Landais, and Søgaard 2019).
The diverging gender wage profiles after the arrival of the first child are often attributed to foregone investments in human capital because of the time out of the labor market (Altonji and Blank 1999). Moreover, a strong switch into working part-time by women following the arrival of the first child can often imply a wage penalty (Manning and Petrongolo 2008). Similarly, mothers avoiding jobs that are often well paid but require overtime or commuting time can also explain part of the gender wage gap that emerges after children join the household (Cortés and Pan 2019; Le Barbanchon, Rathelot, and Roulet 2021).
What has received less attention, however, is the importance of unpredictable (temporary) work absence—or conversely, the importance of worker “presenteeism.” While work absence due to parental leave and part-time employment, which are often prearranged, allows employers to anticipate the absence of a worker, absence due to own sick leave or leave to take care of sick children can be more complicated. These unpredictable absences are often difficult for firms (and researchers) to measure but can, nonetheless, pose important problems for firms.
A firm’s production loss caused by temporary work absence is likely to depend on the internal substitutability of workers, such that presenteeism and its related incentives are likely to be more important in some occupations or firms than in others (Hensvik and Rosenqvist 2019). The more coworkers in the workplace who can easily cover for an absent employee and can satisfactorily perform tasks in the short run in the case of absence, the higher the temporal flexibility. Production disruption and discontinuous drops in productivity are likely to occur in “unique” jobs—that is, if there are no or few close substitutes to the absent employee at the workplace. The importance of presenteeism for a given position suggests that flexibility, such as the possibility of staying at home with a sick child, is more difficult to accommodate.
We first document, using Swedish register data, that both worker presenteeism and job uniqueness are highly rewarded in the labor market. Unique jobs are higher paying, and, at the same time, they provide higher rewards for presenteeism. This continues to be so even after conditioning on occupation and workplace fixed effects. Next, we adopt an event study approach based on changes in within-couple gaps around the birth of the first child. We document the emergence of gender gaps in the probability of holding a unique job after the arrival of the first child, and its consequences for the gender wage gap. By examining changes in temporary absence (or, inversely, presenteeism) before and after the arrival of the first child and connecting it to the premiums for holding a unique job, we highlight the importance of organizational structure on the dynamics of the gender wage gap.
The strong interaction between worker presenteeism and job uniqueness favors more men than women since, after childbearing, women’s presence involves more uncertainty, due to a higher incidence of temporary work absence, and they are more likely to switch into (lower-paying) nonunique jobs.1 Compared with the period before the arrival of the first child, in the relatively short period after, parenthood reduces mothers’ likelihood of holding unique jobs relative to fathers by approximately 3.5 percentage points. This effect grows over time. By approximately 15 years after the child’s arrival, the parenthood effect is approximately six percentage points. We show that these parenthood effects on holding a unique job are not driven by differences in seniority (that is, holding a managerial position) but instead reflect differences across occupations and firms. We connect the premiums for holding a unique job and being present at the workplace with the dynamics of the gender gap in holding a unique job, showing how highly consequential these premiums are for explaining the diverging wage paths of fathers and mothers. We show that unique jobs not only are higher paying but also offer a better wage trajectory and more stable employment.
We further find that the effect of parenthood on the gender gap in holding a unique job is much more (less) pronounced in occupations where the penalty for absence is higher (lower), suggesting gender differences in sorting into and out of jobs that compensate differentially based on the required amount of presenteeism. Women who are initially in occupations where uniqueness is attractive (that is, the uniqueness premium outweighs the penalty of high absence) are less likely to avoid unique positions after childbirth than women who are initially in occupations where uniqueness is unattractive in the event of high absence. Consequently, this implies a smaller gap in the presence in unique positions of fathers relative to mothers. Overall, the results highlight that sorting away from these jobs is likely to contribute to the persistent gender wage gap following the arrival of the first child.
Our study highlights that the internal organization of the firm can play an immensely important role on the gender gap instigated by parenthood. This is consistent with the theoretical framework outlined by Goldin (2014), in which jobs are characterized by their degree of temporal flexibility, where the term temporal flexibility captures how sensitive the productivity in the job is to, for instance, the number of hours worked, and, in particular, the potential to work part-time.2 Our findings emphasize the importance of other dimensions, particularly the sensitivity of the productivity in a job to temporary unpredictable absence (or, alternatively, the importance of presenteeism). We focus on substitutability—namely, when one works in a position with few or no close substitutes at the workplace and therefore with low temporal flexibility—allowing us to explore variation within as well as across occupations.
Our findings underline an important policy implication. It is traditionally believed that increased time out of the labor market after childbearing generates a gender differential in human capital accumulation, which can be an important explanation for the gender wage gap that develops after childbearing (Angelov, Johansson, and Lindahl 2016). In this case, policies aimed at more gender-equal divisions of childcare that reduce gaps in human capital accumulation (for example, paternal quotas) can close the gender wage gap. Here, we highlight that the weight firms attach to presenteeism can be another important channel.
An interesting aspect of the recent crisis caused by the COVID-19 global pandemic is that it has—at least temporarily—decreased the presenteeism requirements of many jobs. There is a great deal of discussion on how this situation has affected the work–life balance within the household, especially when combined with school closures (Del Boca et al. 2020). Women seem to be more adversely affected than men because of an increase in household chores and caring for children but also because of larger employment losses for women than for men (see Alon et al 2020). In the longer run, however, this crisis might actually help to reduce gender inequality.
As a response to the changes in the norms (and needs) of individuals working from home, many firms have invested a great deal in their internal reorganization to accommodate working from home (Barrero, Bloom, and Davis 2020). While this reorganization is likely to have come at a cost, by targeting the organizations’ structure in such a way that tasks are performed in a satisfactory way from home, it can make it less costly for those holding a unique job to not be present in the workplace.
II. Context and Data Description
A. The Swedish Context
From an international perspective, gender equality in the labor market is high in Sweden. The employment and labor force participation rates for women in Sweden are among the highest in the world, and there are relativity small employment differences between men and women. In 2017, the labor force participation rates for women and men aged 20–64 were 85 percent and 89 percent, respectively. The corresponding employment rates were 80 percent and 84 percent, respectively.3 The individualized income taxation and generous public provision of childcare that began in the 1970s have arguably contributed to this situation (Selin 2014). Following a sample of children born in 2004–2009, Hall and Lindahl (2018) report that 55 percent of the children attended formal childcare by 18 months of age. At three years of age, the corresponding number was 93 percent. Aspects of the parental leave system also give incentives for labor force participation before parenthood and between births. Since the parental leave benefits are proportional to foregone wages, women have strong incentives to have a high labor supply in the years immediately preceding parenthood.
Despite the high gender equality in the labor market in Sweden, however, there are still substantial and important differences. The raw gender wage gap (monthly full-time equivalent wages) in 2017 was 11.3 percent. Approximately half of the gap is closed when occupation, sector, education, age, and contracted hours are taken into account (Swedish National Mediation Office 2018), reflecting labor market segregation by gender, which is also common in other countries. Moreover, in 2017, the percentage of women aged 20–64 with full-time employment was only 57 percent, while the corresponding number for men was 74 percent. Thus, the prevalence of part-time work is much higher among women than among men.
With respect to absence, women are also much more likely to use parental insurance than men. In 2017, women took approximately 72 percent of the paid parental leave days and 62 percent of the paid temporary parental leave days—that is, days when the parents abstain from work to care for sick children who would otherwise be in school or in daycare (more details are given in Section II.B). Women also account for approximately 65 percent of the paid sick leave days. Among women in the age group 30–49, many of whom have small children, the corresponding number is almost 70 percent (Statistics Sweden 2018).
Overall, even though the extensive margin labor supply is relatively similar for men and women in Sweden, there are still substantial differences on the intensive margin, in particular during childbearing ages, and with respect to long-run labor market outcomes.
B. Parental Leave Rules in Sweden
In what follows, we will estimate the effect of parenthood on the probability of holding jobs that are more or less substitutable (Section II.C.1 explains in further detail how substitutability is measured). It is, therefore, important to understand the parental insurance system in Sweden. Since 2002, parents who have a child are entitled to 480 paid parental leave days, paid by the Swedish Social Insurance Agency (SSIA). Sixty days are reserved for each parent, and the rest can be divided freely within the couple.4 Over the period of our study, the leave had to be used before the child turned eight. Approximately 75 percent of these 480 paid parental leave days are used during the child’s first two years (Hall and Lindahl 2018).5
During the child’s first 18 months, both the mother and father are allowed to be on full-time parental leave with job protection. The parents are free to choose how many paid parental leave days they want to take during this period. After 18 months, however, the right to parental leave is conditional on taking out parental benefits for the days abstained from work to take care of the child. Therefore, parents are often on unpaid leave for a substantial number of days during the first 18 months, so that they have paid days left to guarantee leave at later ages.
Typically, both the mother and the father are employed full-time in the year preceding the birth of their first child (Angelov, Johansson, and Lindahl 2020). The mother often goes on leave approximately one month before expected delivery and has some income during (at least parts of) this pre-birth leave period (via paid vacation days, paid sick leave days, or paid parental leave days).6 Immediately after the birth of the child, the father is entitled to ten paid temporary parental leave days. These ten days are not part of the 480 days discussed above. Approximately 75 percent of fathers use these ten days and then return to full-time employment, while the mother stays at home with the child (Swedish Social Insurance Agency 2019). Most mothers take full-time parental leave during the child’s first 12 months. The mother takes out paid parental leave days corresponding to approximately 60 percent of this period (Swedish Social Insurance Agency 2013). After 12 months, the mother returns to work, and the father goes on full-time parental leave. The typical father takes full-time parental leave for approximately three to six months (also using paid days for approximately 60 percent of the time). As noted above, parents often use a combination of paid and unpaid leave during the first 18 months, since remaining days with paid leave can be used to guarantee leave at later ages. After this, the child starts attending subsidized daycare.7 At this point, many mothers make at least some reduction in their working hours.8
Once the child is in daycare, and the parents are back in employment, the system of temporary parental leave days becomes relevant. When the child is too sick to be in daycare, the parents have the right to be on temporary parental leave from their job to care for their sick child (only one parent can be at home with the sick child). The parents can then claim temporary parental leave benefits from the SSIA to compensate for their foregone earnings. As in the “standard” parental leave system, the replacement rate is approximately 80 percent up to a cap. Parents are entitled to 120 paid temporary parental leave days annually per child, and the days can be used until the child turns 12 years old. It has previously been shown that the average mother (father) is absent from work to care for a sick child for five (2.5) days per year during the first ten years of the child’s life (Boye 2015; Ichino et al. 2019).9 In Online Appendix Figure A2 (Panel B), we show that women take more temporary parental leave than men in our estimation sample too.
C. Data and Variable Descriptions
Our analysis is conducted using Swedish register data covering (almost) the entire Swedish population between 1997 and 2013.10 We primarily use three different data sources that we link on the individual level via personal identifiers. We briefly discuss the three data sources below.
1. Wages, hours, occupations, and job substitutability
The basis of our analysis is the Wage and Salary Structure Data (WSSD), which contain information on monthly full-time equivalent wages, the extent of the contract (in percent of full-time), and occupations (ISCO-88, three-digit level), as well as worker and establishment identifiers measured in September or November each year.11 The data are collected by Statistics Sweden and cover a representative sample of establishments in the private sector (the data cover almost 50 percent of private sector workers) and all public establishments. The sample of private establishments is stratified by firm size and industry, where establishments within large firms are overrepresented.
Following Hensvik and Rosenqvist (2019), we define employee substitutability as the number of other workers within the same combination of establishment and occupation (ISCO-88, three-digit level) in a given year.12 For example, an office secretary at an establishment that employs five office secretaries will have four substitutes. The data only contain individuals who were employed at the workplace at the point of the survey; that is, individuals who were employed in the same occupation during other parts of the year are not counted as substitutes.13 We construct an indicator variable that takes the value one if the number of substitutes is five or fewer and zero otherwise.14 In what follows, we will use the terms unique (value one) and nonunique (value zero) jobs to describe this distinction. This binary uniqueness variable, and particularly the within-couple difference in this variable, is the main outcome in our analysis. In Online Appendix Table A1, we show characteristics of the employees holding unique and nonunique jobs. Approximately 20 percent of all jobs can be described as unique jobs according to our definition. Employees in unique jobs are, on average, slightly older and more likely to be women than employees in nonunique jobs. Employees in unique jobs also work in more skilled professions and earn higher wages. They also receive less parental and sickness benefits, indicating lower absence.15
2. Benefits from the social insurance system
Using data from the longitudinal database about education, income, and employment (LOUISE), we can observe the amount of sickness absence benefits, parental benefits and temporary parental benefits received from the SSIA on a yearly basis. LOUISE data cover all individuals in Sweden aged 16–74 in a given year (16–65 before 2001).
Sweden has an obligatory, general, and uniform sickness insurance system. Sickness absence benefits are income related, and the replacement rate is almost 80 percent of the labor income up to a cap. Employers pay benefits for the first 14 days of sick leave. Thereafter, the SSIA pays benefits. Individuals in the LOUISE data with a positive value on sickness absence benefits have had at least one spell of sickness absence longer than 14 days during the given year. Among the individuals who have not received sickness absence benefits from the SSIA, there is presumably still substantial variation in actual sickness absence. Some of these individuals have not been absent at all, while others might have had multiple short sickness absence spells (no longer than 14 days). The data do not allow us to observe this variation.
As explained in Section II.B, Swedish parents with children aged eight months to 12 years can also receive temporary parental benefits when they abstain from work to care for sick children who would otherwise be in school or in daycare. The benefits are paid by the SSIA from day one. Thus, unlike the sickness absence benefits, the temporary parental benefits data also pick up short absence spells.
Note that since the benefits are income related, the amount received from the SSIA is a function of both the wage level and the extent of absence. Thus, for example, if men and women are absent to exactly the same extent, men will still receive more absence benefits, on average, since they tend to have higher wages than women.
3. Couples
We draw data on couples from a multigenerational register that covers all individuals in Sweden born 1932–2017. This data set contains the year and month of birth, gender, and a personal identifier that can be linked to other registers. It also contains personal identifiers and birth year for the father and the mother (if they are known). We keep children born in the period 1999–2007, and we require that both the father and the mother are known. The period 1999–2007 is chosen because information on the occupation and thereby the job uniqueness of the parents is available between 1997 and 2013, and we must be able to observe the job uniqueness of the parents both before and after the birth of the child. We further require that the child is the first child for both the father and the mother. A mother and a father who have their first child together are defined as a couple in our data (as in Angelov, Johansson, and Lindahl 2016). We make no further restrictions on the relationship status of the parents.16
In our empirical model, we control for the within-couple difference in job uniqueness two years before the arrival of the first child, and consequently, we further restrict our sample to couples where this difference can be defined—that is, couples in which both the mother and the father are sampled in the WSSD two years before the arrival of the first child. The within-couple difference in job uniqueness is calculated as the father’s value on the job uniqueness variable (one if unique and zero otherwise) minus the mother’s corresponding value.17 This leaves us with 51,729 unique couples. From these couples, we construct a panel of within-couple differences in job uniqueness (and sickness absence) covering the calendar years 1997–2013. However, not all couples are observed each year, since both the father and the mother must be sampled in the WSSD for a given year; that is, it is an unbalanced panel. By taking the difference between the observation year and the birth year of the child, we can construct an event year variable. For example, if a couple is observed in 2003, and they had their first child in 1999, the event year variable will take the value four (2003–1999). In our main analysis, we keep observations where the event year variable takes values in the interval −4 to 14. We then create dummy variables for all event years and all calendar years. For all couples, we also have information on the age difference and the difference in years of schooling two years before the arrival of the first child.
Our final analysis data contain 360,510 observations and 51,729 unique couples. For each observation, we have the following information: the within-couple difference in job uniqueness (and absence), the calendar year of the observation, the event year of the observation, a set of dummy variables for the calendar year of the observation, a set of dummy variables for the event year of the observation, the within-couple difference in job uniqueness (and absence) two years before the birth of the first child, the within-couple difference in age, and the within-couple difference in years of schooling two years before the arrival of the first child.
III. Job Uniqueness and Presenteeism
In this section, we motivate why workers in unique jobs are expected to have higher wage than comparable workers with more substitutes. We then turn to the importance of presenteeism at the workplace and its interaction with job uniqueness.
A. Job Uniqueness and Wages
An inherent feature of temporary and unpredictable work absence is that it should be more costly for firms to find external than internal replacement in the short run. In the rent-sharing model originally outlined by Stole and Zwiebel (1996a,b) and used in Jäger and Heining (2019), the number of substitutes in the firm will therefore affect wages negatively.18 More coworker substitutes could also be regarded as a positive nonwage amenity of the job that employees might be willing to pay for (Rosen 1986).
In line with these arguments, Table 1 shows empirical correlations between job substitutability and wages. We follow the procedure outlined by Goldin (2014) and Cortés and Pan (2019), who estimate the return to long hours in different occupations. Specifically, we restrict the sample to male workers and estimate the following regression:
1where the outcome is the monthly wage of individual i in year t in occupation o at firm f.19 U is a dummy variable indicating whether the individual holds a unique job (that is, takes the value one if the number of substitutes is less than or equal to five, and zero otherwise; see Section II.C.1 for more details). X is a vector of controls that includes a quadratic in age, years of education, and the number of children living at home. We also include year, occupation, and firm fixed effects, as well the interactions between workplace and occupation fixed effects. β reflects the conditional association between the wage and holding a unique job.
Association Between Job Substitutability and Wage—Private Sector Male Employees
From Table 1, we see that there is a strong and positive association between wages and holding a unique job across specifications.20 While this is in part explained by unobservable differences between workplaces and occupations, accounting for these still leaves us with a significantly positive association between job uniqueness and wage. In the most conservative estimation (with individual characteristics and the interaction between workplace and occupation fixed effects, respectively), we see that holding a unique job is associated with around a 2.3 percentage point higher wage. In Online Appendix Figure A1, when we look at different definitions of uniqueness (that is, from zero substitutes to 50, with more than 50 as the reference category), we see that the effect is monotonically decreasing. When there are no other substitutes, the associated wage premium is around five percentage points. After around ten substitutes, the effect is smaller and flatter, until becoming very small and insignificant when the number of substitutes is larger than 25.
The results highlight that the low substitutability of a job is strongly linked to the occupation. However, unique jobs are not only characterized by certain occupations, suggesting that for a given occupation, across firms, there exist different degrees of uniqueness.21
In Online Appendix Table A6, we show the importance of uniqueness in other aspects of the labor market. Employees who hold unique positions are more likely to remain employed in the future and have steeper age profile of wages.22
B. Unique Jobs and Presenteeism
Next, we assess how job uniqueness interacts with presenteeism at the workplace (or, conversely, the importance of unpredictable, temporary work absence). If a firm’s production loss caused by temporary work absence depends on the internal substitutability of workers, presenteeism will be important because “flexibility,” such as caring for sick children, becomes difficult to accommodate when there is no suitable substitute in the workplace.
In the first two columns of Panel A of Table 2, we estimate the relationship between absences due to sick children and wages. Since parents are entitled to sickness benefits for foregone earnings due to absence caused by a child’s sickness, we can separately measure absence due to caring for a sick child.23 There is a strong negative relationship between absence and wages conditional on year, workplace, occupation fixed effects, and covariates, respectively. Thus, those who have been absent to care for sick children in the last year have significantly lower wages.
The Association Between Absence and Wage—Private Sector Male Employees
In the last two columns, we investigate whether the negative relation between absence due to sick children and wages is more pronounced in jobs with low substitutability. This is indeed the case, as indicated by the negative interaction between the absence measure and the uniqueness dummy.24 Moreover, as previously shown in Hensvik and Rosenqvist (2019), employees with few close substitutes in the workplace have significantly lower absence rates than other employees (Online Appendix Table A8), further strengthening the notion that absence is more heavily penalized in jobs with low substitutability.
In Panel B, we replicate the analysis when replacing temporary parental leave with own sickness absence. As noted in Section II, this measure picks up spells of sickness absence longer than 14 days and may thus contain absences that are possibly easier to foresee by employers. Reassuringly, however, we find a negative relationship between own absence and wages that is substantially more pronounced when the number of internal job substitutes is lower.25
IV. Parenthood and Job Uniqueness
A. Event Study Approach
To explore changes in the probability of holding a unique job, we adopt an event study approach based on changes around the birth of the first child for a mother relative to a father. This event study methodology closely follows Angelov, Johansson, and Lindahl (2016), who compare the income and wage trajectories of women to those of their male partners before and after parenthood.
We follow couples (indexed by i) before and after the arrival of the first child (indexed by t). We study couples who have their first child over the period 1999–2007, following them until 2013 (calendar years are indexed by c). In the regression analysis, we restrict the sample to couples in which both the mother and the father are observed in the WSSD two years before birth, so that we can compute the within-couple gap in job uniqueness at t = −2 (51,729 couples). Both the father and the mother must be observed in the WSSD at a certain event time for the within-couple gap in job uniqueness to be defined at that event time.26
The empirical model is specified as:
2The outcome is the within-couple gap in job uniqueness (father–mother) at a particular event time (t), which ranges from four years before birth to 14 years after birth. The outcome is explained by event time dummies (where t = −1 is the omitted category), calendar year dummies (where c = 1997 is the omitted category), the within-couple gap in job uniqueness at t = −2 (the lagged y-variable), the within-couple differences in age and pre-birth years of education (x variables), and an error term. We are interested in the coefficients on the event time dummies (αj), which, conditional on the model, identify the effect of parenthood on the change in the within-couple gap in job uniqueness relative to the pre-birth difference.
When linking the analysis around presenteeism to the gender gap in holding a unique job following the birth of the first child, it is quite likely that the certainty around being present at the workplace might be altered by the presence of children. Unlike periods of maternity leave and lower (predicted) labor supply (for example, working part-time), temporary absence, often due to sickness, is difficult to measure or account for—especially if the absence is related to caring for sick children. In Online Appendix Figure A2, we replicate the findings of Angelov, Johansson, and Lindahl (2020) to show that within our sample too, mothers substantially increase their sick leave compared with fathers after childbearing. In Panel A, we plot αj coefficients from Equation 2 when we have replaced the within-couple gap in job uniqueness as the outcome with the within-couple gap in own sick leave benefits. Consistent with earlier work, there is a significant gender gap in sickness absence that emerges at the onset of childbearing and that does not dissipate over time. In Panel B, we show that, consistent with findings of Boye (2015) and Ichino et al. (2019), mothers are also more likely to take temporary parental leave absence from work compared to their male partners.27
Figure 1 plots estimates of αj in Equation 2 when the model is estimated on our baseline sample of couples, showing that the pre-birth event time dummies are generally small and nonsignificant. The coefficients on the post-birth event time dummies, on the other hand, are significantly positive and large, as well as persistent.28 The results for the baseline sample clearly show that men have an increased likelihood of holding jobs with few substitutes relative to women after the arrival of the first child.29 Three years after childbirth, the within-couple gap in job uniqueness has increased by 3.4 percentage points relative to the pre-birth difference. In the long run, this parenthood effect increases and persists, with long-run parenthood effects of approximately five to six percentage points. Online Appendix Table A9 presents the estimates.30
Effect of Parenthood on the Within-Couple Gap in Uniqueness: Baseline Sample
Notes: The figure shows estimates of αj in Equation 2, together with 95 percent confidence intervals, for the baseline estimation sample. In Equation 2, the outcome is the within-couple gap in uniqueness (father–mother) at a particular event time, which ranges from four years before birth to 14 years after birth. Uniqueness is defined as having less than or equal to five coworkers in the same occupation. Uniqueness can be observed between 1997–2013. The outcome is explained by event time dummies (where t = −1 is the omitted category), calendar year dummies (where c = 1997 is the omitted category), the within-couple gap in uniqueness at t = −2, the within-couple differences in age and pre-birth years of education, and an error term. The coefficients on the event time dummies (αj) identify the effect of parenthood on the change in the within-couple gap in uniqueness relative to the pre-birth difference. In the baseline estimation sample, we include all couples with nonmissing values on the within-couple gap in uniqueness two years before the birth who had their first child over the period 1999–2007 (51,729 unique couples). The number of observations in the baseline estimation is 360,510.
The results highlight that presenteeism-related costs create a substitutability relationship between having children and holding unique jobs. Although substitutability is important, regardless of the direction of the effect, the event study approach tries to isolate the impact of having a first child on future job uniqueness. While it is difficult to completely rule out anticipated effects (that is, couples deciding to have children in response to an anticipated future uniqueness gap increase), the event study is constructed to reflect mostly the effects from children on the uniqueness gap. By showing that there are no pre-trends in the uniqueness gap within couples, we show, on average, previous changes in the uniqueness gap do not predict having children. Moreover, changes in job uniqueness are quite sharp after maternity leave for the first child ends.
Most couples (84 percent) in our sample have at least one additional child within seven years after the arrival of their first child. In which case, it is likely that the arrival of additional children plays a role in the dynamic patterns. While the first-year effect is partially dampened by maternity leave, the within-couple uniqueness gap grows sharply two years after having a child. While additional children are potentially endogenous to job uniqueness, we explore the effects by number (and spacing) of additional children in Online Appendix Figure A5. The results indicate that the effect of the first child is present and lasting. Moreover, the additional gap in uniqueness over time is indeed driven by couples who have additional children during the follow-up period, whereby having more than one child, conditional on having at least one, is the norm in the data.
In Online Appendix Figure A6, we further investigate to what extent the effects are driven by women. We see that, in the first years after the arrival of their first child, the effect is largely driven by mothers switching out of a unique job. Fathers, on the other hand, do not change their behavior, and we see that, over time, there is an increase in their probability of holding a unique job (which exacerbates the gap within the couple in the longer run).
B. Job Uniqueness and Seniority
Could holding a unique job simply reflect holding a more senior (or managerial) position? This could be the case, especially in later years, and could, therefore, reflect the gender gap in holding a manager position. In Figure 2, we show that when excluding managers, while the effects are smaller than those in the baseline model, sizeable effects are still present.
Effect of Parenthood on the Within-Couple Gap in Uniqueness: Other Samples
Notes: See note in Figure 1. In the “private sector” estimation sample, we restrict the analysis to jobs in the private sector. The “private sector” estimation sample includes 105,033 observations (17,997 unique couples). In the “public sector” estimation sample, we restrict the analysis to jobs in the public sector. The “public sector” estimation sample includes 91,386 observations (12,372 unique couples). In the “nonmanagers” estimation sample, we drop observations where either the father or the mother holds a managerial position. The “nonmanagers” estimation sample includes 319,830 observations (48,899 unique couples). In the “low-skilled” estimation sample, we restrict the analysis to jobs in low-skilled occupations. The “low-skilled” estimation sample includes 67,398 observations (13,992 unique couples). In the “high-skilled” estimation sample, we restrict the analysis to jobs in high-skilled occupations. The “high-skilled” estimation sample includes 171,944 observations (23,563 unique couples). In the “same occupation” estimation sample, we keep couples in which the mother and the father worked in the same occupation two years before birth. The “same occupation” estimation sample includes 65,764 observations (8,549 unique couples). In the “different occupation” estimation sample, we keep couples in which the mother and the father worked in different occupations two years before birth. The “different occupation” estimation sample includes 294,746 observations (43,180 unique couples).
Importantly, we also show that the baseline results are robust to restricting the sample to couples in which the mother and the father worked in the same occupation two years before birth, suggesting that the baseline results are not driven by men and women working in different pre-birth occupations with potentially different time profiles with respect to job uniqueness. Figure 2 further shows that the estimates remain similar when we restrict the analysis to couples working in the private sector or in the public sector, as well as in low- versus high-skilled jobs, suggesting no difference in the gender gap when in one sector or the other.31 Online Appendix Table A9 reports the estimates.32
In sum, the patterns that we uncover are present generally in unique jobs and not just managerial or higher-skilled positions. The effects are driven by differences across firms in the number of substitutes for a given occupation and the penalty associated with absence. In other words, the more coworkers in the workplace who can easily cover for an absent employee and can satisfactorily perform tasks in the short run in the case of absence, the higher the temporal flexibility.
C. Role of Absence-Sensitivity of Pre-Birth Job
If women avoid jobs with few substitutes after the arrival of the first child because they would otherwise incur severe absence-related wage cuts, we would expect more pronounced effects of parenthood in occupations where jobs with few substitutes are particularly sensitive to absence. In this section, we investigate the relevance of this hypothesis.
Jobs with low substitutability appear to have a wage premium (Table 1), but at the same time, absence is more heavily penalized in such jobs (Table 2). Employees who anticipate a high absence rate must reflect on this trade-off and the net effect of the premium versus the extra absence penalty when making a decision about staying in (or leaving) jobs with few substitutes. This net effect on the wage might, of course, vary by occupation. In some occupations, it can be relatively attractive to have a job with few substitutes, even in the presence of a high absence penalty. However, in other occupations, it may be that the cost outweighs the benefit.
We can estimate the occupation-specific net effects on the wage of having a unique job but exhibiting high absence by running versions of the model used in Table 2 separately by occupation and summing the coefficients on “Unique job” and the interaction. We estimate the occupation-specific net effects by year and then generate a dummy variable indicating whether the occupation is above the median in terms of the net effect in a given year. If the occupation is above the median, it can be interpreted as a relatively “attractive” job with few substitutes in that occupation, even if the individual expects to have high absence. If the occupation is below the median, it can be interpreted as a relatively “unattractive” job with few substitutes in that occupation if the individual expects to have high absence.33
Following this procedure, we find that jobs in the service sector (personal care and related workers, primary education teaching professionals, and nurses) are among the top “attractive” jobs (for both male and female employees). Among “unattractive” jobs, we find jobs, such as, archivists, librarians and related information professionals, and ship and aircraft controllers and technicians.
Since we know the pre-birth occupation (the occupation two years before birth) of all mothers in our couple sample and the calendar year in which the pre-birth occupation was observed, we can match the occupation and year to add information about the attractiveness of holding a unique job in the mother’s pre-birth occupation. Therefore, two years before they have a child, we know whether the women work in an occupation where job uniqueness is attractive even in the case of high temporary absence (due to positive net wage effects). Women in pre-birth occupations where uniqueness is attractive should be less likely to avoid unique positions after childbirth than women who are in pre-birth occupations where uniqueness is unattractive in the case of high absence. Consequently, in couples where before the birth of the first child, the woman works in an occupation where job uniqueness is attractive even with high absence, the effect of parenthood on the within-couple gap in holding a unique position should be smaller than that in couples where the mothers work in an occupation where uniqueness is unattractive (for a given level of high absence).
We estimate the model in Equation 2 separately for these two couple types and plot the point estimates of the αj parameters in Figure 3. Consistent with the prediction, we find substantially larger effects of parenthood on the within-couple gap in uniqueness in couples where the mother was in a pre-birth occupation where uniqueness was unattractive in the case of high absence. In addition, when we replace uniqueness as the outcome with the log wage, we find a similar pattern—that is, the effect of parenthood on the within-couple gap in wages is larger in couples where the mother was in a pre-birth occupation where uniqueness was unattractive in the case of high absence (Figure 4). While these results should be interpreted with some caution, they suggest that women sort away from jobs where absence is heavily penalized after the onset of parenthood. Online Appendix Table A10 shows the estimates in Figures 3 and 4.
Effect of Parenthood on the Within-Couple Gap in Uniqueness
Notes: The figure shows estimates of αj in Equation 2 for an unbalanced panel of couples divided into two groups depending on the net cost of absence related to low substitutability. In Equation 2, the outcome is the within-couple gap in uniqueness (father–mother) at a particular event time, which ranges from four years before birth to 14 years after birth. Uniqueness is defined as having less than or equal to five coworkers in the same occupation. Uniqueness can be observed between 1997 and 2013. The outcome is explained by event time dummies (where t = −1 is the omitted category), calendar year dummies (where c = 1997 is the omitted category), the within-couple gap in uniqueness at t = −2, the within-couple differences in age and pre-birth years of education, and an error term. The coefficients on the event time dummies (αj) identify the effect of parenthood on the change in the within-couple gap in uniqueness relative to the pre-birth difference. We divide our baseline estimation sample (that is, all couples with nonmissing values on the within-couple gap in uniqueness two years before the birth who had their first child over the period 1999–2007) into two parts: couples in which the mother had a pre-birth occupation where uniqueness is relatively attractive even with high absence and couples in which the mother had a pre-birth occupation where uniqueness is relatively unattractive in the case of high absence. This division is further explained in the main text of Section IV.C. The “attractive” estimation sample includes 199,871 observations (29,030 unique couples). The “unattractive” estimation sample includes 154,739 observations (22,049 unique couples).
Effect of Parenthood on the Within-Couple Gap in Log Wage
Notes: The figure shows estimates of αj in Equation 2 for an unbalanced panel of couples divided into two groups depending on the net cost of absence related to low substitutability. The outcome is now the within-couple gap in log wage (father–mother) at a particular event time, which ranges from four years before birth to 14 years after birth. The outcome is explained by event time dummies (where t = −1 is the omitted category), calendar year dummies (where c = 1997 is the omitted category), the within-couple gap in log wage at t = −2, the within-couple differences in age and pre-birth years of education and an error term. The coefficients on the event time dummies (αj) identify the effect of parenthood on the change in the within-couple gap in log wage relative to the pre-birth difference. We divide our baseline estimation sample (that is, all couples with nonmissing values on the within-couple gap in log wage two years before the birth who had their first child over the period 1999–2007) into two parts: couples in which the mother had a pre-birth occupation where uniqueness is relatively attractive even with high absence and couples in which the mother had a pre-birth occupation where uniqueness is relatively unattractive in the case of high absence. This division is further explained in the main text of Section IV.C. The “attractive” estimation sample includes 199,871 observations (29,030 unique couples). The “unattractive” estimation sample includes 154,739 observations (22,049 unique couples).
V. Discussion
We have shown that temporary unpredictable absence is penalized more heavily in jobs where there is a scarcity of close substitutes at the workplace. Consistent with this, we have shown that the dramatic rise in women’s temporary absence rates, both due to own sickness and due to care for sick children, following the arrival of the first child is accompanied by a sharp and persistent drop in the likelihood of holding jobs with few substitutes. Recent studies have shown that women tend to sort into more family-friendly sectors and workplaces after the arrival of the first child (Hotz, Johansson, and Karimi 2018; Kleven, Landais, and Søgaard 2019). Here we use information at the job level to construct a within-firm-occupation measure of worker substitutability. We then show that women tend to sort into more family-friendly positions with more substitutability.
Theoretically, our findings are consistent with the framework outlined by Goldin (2014), specifically with a model in which jobs are characterized by their degree of temporal flexibility, where the term temporal flexibility captures how sensitive the productivity in the job is to, for instance, the number of hours worked and the precise hours worked. While Goldin’s discussion of temporal flexibility in different jobs largely centers around the potential productivity drops associated with working part-time, temporal flexibility can encompass many more dimensions. Our study highlights the importance of other dimensions, particularly the sensitivity of the productivity in a job to temporary unpredictable absence (or, alternatively, the importance of presenteeism). Incorporating temporary unpredictable absence into the discussion of the temporal flexibility of a job offers a direct correspondence between the empirical analysis of our study and the model described by Goldin (2014), which highlights the generalizability and usefulness of the model.
In Goldin’s model, the productivity of a worker in a job with low temporal flexibility drops discontinuously when “the worker is absent more than some amount.” In our interpretation of the model, this discontinuous drop in productivity can occur if the employee works too few hours overall and/or if the temporary unpredictable absence of the employee becomes too common. In jobs with high temporal flexibility, workers can be absent much more before productivity drops, and the drop might be less pronounced. Goldin (2014) postulates a link between the temporal flexibility of a job and the substitutability among workers on the workplace. Stated simply, the more coworkers in the workplace who can easily cover for an absent employee and satisfactorily perform tasks in the short run in the case of absence (that is, the closer substitutes there are), the higher the temporal flexibility of a job. If there are no or few close internal substitutes for the absent worker, the consequences of temporary unpredictable absence can be costly production disruptions or cancelled meetings with important clients.
In the model, jobs with low temporal flexibility—in our case, jobs with few substitutes—pay a high wage when the labor supply requirement is met (that is, when absence is low and predictable) and a low wage otherwise (that is, in these jobs, the wage penalty of absence is high). Jobs with higher temporal flexibility pay a lower baseline wage, but on the other hand, the wage penalty of absence is lower. Thus, for some individuals, it will be optimal to transition into a job with higher temporal flexibility when the rates of temporary unpredictable absence increase. In the same way, the incentives to transition into a job with low temporal flexibility will be weaker when the rates of temporary unpredictable absence increase. Since women increase their rates of temporary unpredictable absence relative to men after the arrival of the first child, the model would predict that parenthood will reduce the relative likelihood of women holding jobs with low temporal flexibility—in our case, jobs with few substitutes. This is what we have empirically identified in the data.
Overall, our results highlight the importance of occupational structure. Goldin (2014) argues that the cost of absence is reduced when tasks and procedures are standardized, and in our study, we contribute to this finding by showing that the cost of absence can also be lower with more balanced occupational groups in the workplace. Our study focuses on one form of substitutability—namely, when one works in a position where there are few or no close substitutes within the workplace. Other forms of frictions related to temporal substitution might constrain employees in a similar way, for instance, working in a position with a great deal of individual autonomy or with incentives based on the employee’s own client list.
Unlike parental leave and part-time employment, which allow the employer to anticipate the absence of the worker, temporal work absence, often due to own sickness or caring for sick children, is unpredictable. This study highlights that presenteeism plays an important role in explaining the parenthood wage penalty for women.34 From the firm’s perspective, it is likely that such reorganization may come at a cost, depending on the specific tasks to be shared. An interesting aspect of the recent COVID-19 crisis is that it has incentivized firms to reorganize work to accommodate social distancing guidelines and increased absence rates caused by the COVID-19 virus. This push towards greater work flexibility could make it less costly for workers to be absent and lower the need for internal replacement. Exploring further the implications of these organizational changes for gender inequality is an important area for future research.
Acknowledgments
The authors thank Erica Lindahl, Astrid Kunze, Håkan Selin, and seminar participants at the Institute for Evaluation of Labour Market and Education Policy (IFAU) in Uppsala, University of Oxford, Dauphine University, University of Konstanz, for their helpful comments and suggestions. The authors declare that they have no relevant or material financial interests that relate to the research described in this paper. The paper relies on restricted-access individual-level administrative data from Sweden. Due to the sensitive nature of these data, they are confidential and protected according to 24 ch. 8§ of the Public Access to Information and Secrecy Act (2009:400). We are therefore prohibited from posting the data online. Researchers can apply for access to the data used in the study from the Institute for Evaluation of Labour Market and Education Policy (IFAU). IFAU decides, after examination pursuant to 24 ch. 8§ of the Public Access to Information and Secrecy Act (2009:400), to disclose the data with a reservation of secrecy in accordance with 10 ch. 14§ of the Public Access to Information and Secrecy Act. Access to the data will to some extent depend on the location of the researcher and must be handled on a case-by-case basis. Contact IFAU (IFAU, Box 513, 75120 Uppsala, Sweden; ifau{at}ifau.uu.se) for more details on the application.
Footnotes
↵1. We can measure changes in temporary absence before and after the arrival of the first child, separately identifying whether the absence is due to caring for a child or for oneself. Consistent with previous studies (Angelov, Johansson, and Lindahl 2020; Boye 2015; Ichino et al. 2019), we show that women are more likely to experience an increase in both types of absence relative to men at the onset of childbearing.
↵2. Goldin (2014), Goldin and Katz (2016), and Bütikofer, Jensen, and Salvanes (2018) present empirical evidence on the child wage penalty for college-educated women across majors, showing that mothers in professions with more nonlinear wage structures, such as law, have a larger penalty than those in professions with a more linear wage structure, such as pharmacy or medicine.
↵3. See Statistics Sweden (2018).
↵4. For children born from January 1, 2016, the corresponding division is: 90, 90, and 210.
↵5. In 2014, the limit was extended to the age of 12. However, with the new age limit, parents are only allowed to save a maximum of 20 percent of the 480 days after the child’s fourth birthday.
↵6. The mother can start taking out paid parental leave days 60 days before expected delivery. However, it is quite rare that mothers actually use these paid days before the birth of the child.
↵7. The monthly cost of having one child in daycare is 3 percent of the household’s pretax monthly income up to a cap. The maximum cost per month is currently (2019) 1,425 SEK (≈168 USD).
↵8. The part-time incidence among working mothers (fathers) with young children is 34.6 (8.9) percent, according to Statistics Sweden (2018).
↵9. The use of temporary leave to care for sick children peaks when the children are between two and seven years old, which corresponds to the time when Swedish children attend childcare (Boye 2015).
↵10. In general, Swedish register data go back to 1985, but information on occupations (and, hence, uniqueness) is available from 1997, which is therefore our starting point.
↵11. Private sector firms are sampled in September, whereas public sector establishments are sampled in November each year.
↵12. There are 107 different occupations in our data at the three-digit level.
↵13. When we look at the data over time, we see that around 70 percent of workers remain in a given workplace*occupation cell in two consecutive years (conditional on that the establishment is sampled in both years), suggesting that most workers are on long-term contracts.
↵14. We follow Hensvik and Rosenqvist (2019) in defining uniqueness as five or fewer substitutes. We complement the analysis, however, by looking also at the full distribution of the number of substitutes. This allows us to check the monotonicity of effect of uniqueness on the outcome of interest and to measure the change in coefficient as the number of substitutes increase.
↵15. In Online Appendix Table A2, we restrict the sample to individuals holding unique jobs and show how the jobs are distributed over occupations on the one-digit level separately for men and women.
↵16. About 94 percent of the parents in our sample live together three years after the birth.
↵17. All within-couple differences defined in the paper are calculated the same way, that is, the father’s value on the relevant variable minus the mother’s value.
↵18. More specifically, assuming decreasing returns to scale and a wage-setting process where the surplus is split equally between the worker and the firm a worker’s wage will be a weighted average of the marginal products integrated over the number of substitutes.
↵19. We restrict attention to male workers when computing the returns to an aspect of a job to overcome potential selection concerns that are likely to affect wages and hours of female workers.
↵20. This is consistent with the results in Hensvik and Rosenqvist (2019).
↵21. In Online Appendix Table A3, we show that the estimates in Table 1 remain significantly positive also after excluding managers from the sample. In Online Appendix Table A4, we further show that the estimates in Table 1 are robust to variations in the cutoff value for uniqueness (that is, 3 or 7, instead of 5). In Online Appendix Table A5, we show that they are robust to the inclusion of partner demographics and uniqueness.
↵22. The effects are fairly large, reducing the nonemployment rate by approximately 0.3 percentage points from a baseline nonemployment rate of 3.4 percent.
↵23. This analysis is restricted to employees with children aged 0–10 since it is for this age band that parents are entitled to take leave to care for sick children.
↵24. In Online Appendix Table A7, we reestimate the models in Panel A of Table 2 using absence measured in days instead of the amount. This transformation requires us to know the wage in t-1 which is available for about 80 percent of the sample. The results are robust to this alternative way of measuring absence.
↵25. In Online Appendix Table A5, we have estimated the models in Column 4 of Table 2 on a subsample of men for whom we can observe and control for partner characteristics. The results look very similar in this specification.
↵26. As discussed in Section II.C.1, the WSSD cover all employees in the public sector but only 50 percent of the employees in the private sector (a representative sample). For a couple to be observed at a certain event time, both the mother and the father must be working at that time, and conditional on working, they must both be sampled. This implies an unbalanced panel of couples (as in Angelov, Johansson, and Lindahl 2016).
↵27. Note that during t = 0 and t = 1 parents typically are on regular parental leave and the absence patterns must therefore be interpreted with caution as they may interact with the standard parental leave system. For instance, in Panel A the small drop at t = −1 and the large drop at t = 0 reflect pregnancy-related sickness absence. The effect close to zero at t = 1 reflects that most women are on parental leave at this point, making them less likely to use the sickness insurance system.
↵28. Controlling for a linear pre-trend we find that the coefficients are somewhat smaller but continue to be positive and highly significant in most years, such that the overall picture remains unchanged (see estimates in Online Appendix Table A9).
↵29. The results are very similar when we restrict the sample to the 94 percent of the couples that cohabitate three years after birth (Online Appendix Figure A3).
↵30. We complement this analysis by looking at a more continuous measure of uniqueness. In Online Appendix Figure A4, we plot the within-couple gap in the adjusted number of substitutes (father–mother) at a particular event time, which ranges from four years before birth to 14 years after birth. We see very similar patterns as using the job uniqueness dummy. Following the arrival of the first child, there is a sharp and significant drop in the relative number of substitutes for fathers than mothers. This gap continues to grow for some years after the arrival of the first child.
↵31. In our baseline sample, because of the nature of the WSSD data, couples where both the mother and the father are working long-term in the public sector are overrepresented, which could pose a challenge for the representativeness of our main results. The similarity of the results for the private and public sector in Figure 2, however, reduces this concern. We have also estimated Equation 2 replacing the within-couple difference in job uniqueness with the within-couple difference in income. Since income, unlike job uniqueness, is available for the entire population each year, we can compare the results on income in our baseline sample with the results on income in a wider and more general sample. It turns out that the results on income are generally similar in the two samples, further supporting the representativeness of our main results. There are, however, some differences in the very long run (t = 12, 13, and 14), which suggests some caution when interpreting the results that are identified from a relatively small set of couples.
↵32. In Online Appendix Figure A7, we show that the results in Figure 2 are robust to variations in the cutoff value for uniqueness (that is, 3 or 7, instead of 5).
↵33. To maximize precision, we rely on the model in Table 2, Panel B (own sickness absence) when deriving the net effects as this allows us to use the full sample and not only parents with small children.
↵34. Besides a move towards jobs with higher substitutability, gender gaps in presenteeism linked to the presence of children may directly impact gender wage gaps through other channels such as direct productivity impacts and statistical discrimination.
- Received November 2021.
- Accepted September 2022.
This open access article is distributed under the terms of the CC-BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) and is freely available online at: https://jhr.uwpress.org.
Jump to section
Related Articles
Cited By...
- No citing articles found.