Fatherhood timing and men’s midlife earnings: A within-family study of Finnish cohorts born in 1938−50

Abstract

Fathers tend to achieve higher earnings than childless men, but there is limited evidence on the associations between fatherhood timing and men’s later earnings. Using a longitudinal census-based sample of Finnish men, including a subsample of brothers, we investigated fatherhood timing and men’s midlife earnings using both between- and within-family models. Earnings around age 50 were lower among adolescent and young fathers than for men who became fathers at ages 25–29 or later, but these associations became negligible after accounting for measured confounders and unobserved familial confounding. Overall, our findings highlight the important roles of selection into early childbearing and into childlessness. At the population level, early fatherhood was associated with clear negative distributional shifts in fathers’ midlife earnings. However, among all men, any influence of fatherhood timing on men’s midlife earnings distribution paled in comparison with that of childlessness.

Keywords:

• fatherhood timing
• early fatherhood
• male fertility
• childbearing
• childlessness
• earnings
• income
• life course
• sibling comparison

Introduction

The gendered consequences of parenthood are a persistent source of economic inequality (Juhn and McCue 2017; Van Winkle and Fasang 2020; Icardi et al. 2022; Nisén et al. 2022). While women’s earnings are known to be negatively affected by motherhood (Waldfogel 1997; Kahn et al. 2014; Glauber 2018), and by early childbearing in particular (Blackburn et al. 1993; Drolet 2002; Miller 2011; Nisén et al. 2022), fathers seem to enjoy a ‘fatherhood premium’ in earnings (Lundberg and Rose 2000; Petersen et al. 2014; Icardi et al. 2022). Despite the earnings advantage of fathers relative to childless men, the adverse effects of early parenthood observed among women may apply to men as well. Although fathers rarely experience the career interruptions due to childbearing that mothers do, parental responsibilities at a young age may influence young men’s educational and employment choices and lead them into lower-paid employment and less stable careers in the long term (Sigle-Rushton 2005). However, studies on fatherhood timing and its relationship with men’s later earnings (Dariotis et al. 2011; Nisén et al. 2022) and economic well-being (Sigle-Rushton 2005; Assini-Meytin and Green 2015) are still scarce.

To the best of our knowledge, no previous study of the long-term economic consequences of fatherhood timing has used within-family comparison to control for the unobserved social and genetic characteristics shared by brothers. Comparing siblings who become fathers at different ages is a valuable approach, as both entry to fatherhood and earnings can be affected by unobserved characteristics that induce bias in the estimation of the effects of fatherhood (Kunze 2020). For example, men who father a child early in life are more likely to come from a disadvantaged background (Michael and Tuma 1985; Kiernan 1997; Woodward et al. 2006; Sipsma et al. 2010), and this could translate into lower educational attainment and later earnings regardless of parenthood and its timing (Geronimus and Korenman 1993; Assini-Meytin and Green 2015). The influence of family background is likely to be particularly important in early career transitions: social ties with parents have been shown to help offspring find their first jobs (Kramarz and Nordström Skans 2014), while ties with acquaintances may provide useful labour market information (Granovetter 1983). Brothers are much more likely to share similar social ties and networks, work-related values, and knowledge of the labour market as compared with individuals selected randomly from a population. Besides shared socio-economic background, there are also well-documented patterns of intergenerational transmission of reproductive behaviour, including the timing of parenthood (Steenhof and Liefbroer 2008). Therefore, the unobserved heterogeneity in genetic endowment, family resources, social networks, and individual values is better taken into account by the within-family approach.

Using a large population-representative sample of Finnish residents in the 1950 Census, the study described in this paper investigated the relationship between fatherhood timing and men’s midlife earnings. The unique data allowed us to identify brothers born in 1938–50 (n  =  21,961) and make use of register linkage to their reproductive histories, socio-demographic characteristics, and later earnings. For these cohorts of men, family formation still followed a mostly traditional pattern, with strong norms of marrying before childbirth and staying married (Finnäs 1995), and parental leave policies had not yet been extended to fathers (Haas and Rostgaard 2011). Despite relatively high employment rates among women in Finland, men provide a larger share of household earnings (Rønsen and Sundström 2002; Nieuwenhuis et al. 2017). Our main hypothesis was that the younger a man was at first childbirth, the lower his midlife earnings would be. To address selection behind the associations between fatherhood timing and later earnings, the study used a within-family comparison that controls for the confounding influences of unobserved characteristics common to brothers, as well as observed differences in their education, health, and demographic characteristics. Finally, we assessed the population-level influence of fatherhood timing on the overall distribution of fathers’ and all men’s midlife earnings.

Background

Fatherhood premium in earnings

Previous studies of the economic consequences of fatherhood have focused mainly on wage differentials between fathers and non-fathers or looked at wage changes before and after having children (Lundberg and Rose 2000; Hodges and Budig 2010; Killewald 2013; Petersen et al. 2014). These studies using panel data have often documented higher wages among fathers (the fatherhood premium), although the magnitudes of the associations appear to vary across time and societal context (Juhn and McCue 2017; Mari 2019; Cooke et al. 2022; Icardi et al. 2022), as well as along the earnings distribution (Cooke 2014; Glauber 2018; Icardi et al. 2022). The presumed mechanisms behind the economic advantage of fatherhood usually fall into the broad theoretical categories of causal or selective effects (i.e. fatherhood inducing changes in men’s behaviour vs selection of men with higher earnings potential into fatherhood), but some approaches also incorporate outside influences such as positive employer discrimination (Mari 2019; Icardi et al. 2022).

Empirically, it has been difficult to discern whether fatherhood encourages men to earn more or whether men with higher earnings potential and productivity are selected into fatherhood (Lundberg and Rose 2002). Traditionally the most common approach—largely based on Becker’s (1981, 1985) theory of specialization in market and non-market labour between spouses—suggests that, normatively, good fathers will increase their breadwinner capacity after the birth of their first child (Hodges and Budig 2010; Killewald 2013). Consistent with the theory, Lundberg and Rose (2000) showed that if wives experienced an interruption in employment after childbirth, fathers’ working hours and wages increased, whereas if wives remained attached to the labour force, fathers’ working hours decreased. Besides increased work effort among men who become fathers (Astone et al. 2010; Gibb et al. 2014), it has also been suggested that in the absence of other information, fatherhood may signal positive attributes such as loyalty, commitment, and productivity, which are highly valued by employers (Hodges and Budig 2010). Yet, empirical evidence on the role of positive employer discrimination behind better labour market outcomes and wage premiums for fathers remains scarce (Correll et al. 2007; Yu and Hara 2021). Positive selection, whether due to apparent or actual characteristics of men who become parents vs those who do not, is nevertheless likely to account for some of the observed associations. In fact, recent studies exploiting within-individual or within-group variation, based on longitudinal data from the UK, Germany, and Finland, have provided little support for any notable causal fatherhood wage premium (Mari 2019; Icardi et al. 2022). In line with these results, a Norwegian register study using a within-family approach found only a small increase in earnings after first birth among brothers and none among twins (Kunze 2020), emphasizing the role of selection into fatherhood.

Fatherhood timing and later earnings

Despite the vast literature on how fatherhood may relate to economic outcomes, empirical research on parenthood timing and earnings has largely ignored men. Although career interruptions due to childbearing are inarguably larger among women, early fatherhood could also disrupt men’s career-building and affect their later earnings. Parental responsibilities may conflict with young men’s investment in education and training, inducing a preference for the immediate returns of short-term education and early employment over higher future returns from academic education, leading them into less stable and lower-paid employment careers (Sigle-Rushton 2005). For example, in a United States (US) study, young fathers were shown to work fewer hours in the long run, despite an initial increase (Weinshenker 2015). Selection into early fatherhood in the first place is, however, likely to be an important factor behind any differences by parenthood timing in men’s later earnings (Sigle-Rushton 2005; Woodward et al. 2006; Assini-Meytin and Green 2015).

In an empirical analysis of disadvantaged neighbourhoods in Chicago, Assini-Meytin and Green (2015) showed that teenage fathers were more likely to be unemployed than non-teenage fathers at age 32, independent of selection into teenage fatherhood (which was taken into account by using propensity score matching techniques). The subsequent socio-economic disadvantage of teenage fathers disappeared by age 42, however (Assini-Meytin and Green 2015). In a similar study from the UK, Sigle-Rushton (2005) indicated that becoming a father before age 22 increased the use of social benefits but was unrelated to later-life occupational status or unemployment. In contrast, Dariotis et al. (2011) indicated that early fatherhood was associated with lower earnings in the US, and the relative disadvantage associated with early fatherhood increased across the life course from ages 27 to 37. In Germany and Finland, postponement of parenthood has also been found to be associated with somewhat higher later income among men (Kind and Kleibrink 2012; Nisén et al. 2022; Vecgaile 2022).

These existing high-quality studies on fatherhood timing have examined mainly the associations between early fatherhood and early midlife economic outcomes (for men aged up to their 40s), with mixed results. The dissimilarities in findings could relate to the different outcomes used, the techniques used to deal with unobserved heterogeneity (which affects the relationship between fatherhood timing and later economic outcomes), the focus on selected subpopulations, and/or the specific societal contexts. To the best of our knowledge, the present study is the first on fatherhood timing and long-term economic outcomes to use within-family comparison to control for selection into early and later fatherhood by shared family background characteristics.

Study hypotheses

Based on previous literature, we expect to find an early fatherhood penalty in men’s midlife earnings. While parenthood might encourage young fathers to provide for their families, early fatherhood may also disrupt long-term career-building and lead to lower earnings in later life. However, we expect any such long-term earnings disadvantage among young fathers to be at least partly due to selection into early fatherhood by family background. We do not put forward any a priori hypothesis on the association between later-than-average fatherhood and midlife earnings: although delaying family formation to gain educational qualifications is likely to increase later earnings, further delays in employment careers or potential employer discrimination against non-fathers could have opposite effects. Finally, we expect to find a clear earnings disadvantage among men with no children that is also likely to be partly explained by family background.

Finnish context

The mean age at first birth among our study sample of brothers born in 1938–50 was 26.2 and the median age 25.0 years, while 19 per cent had no children by age 52 (the same as for our sample of all men, not just brothers). The mean number of children among all men in these birth cohorts was 1.77, with a standard deviation of 1.33 (Einiö et al. 2016). Until the late 1960s, childbearing occurred almost exclusively within marriage in Finland, and non-marital cohabiting relationships existed mainly as a short prelude to marriage (Finnäs 1995). Births outside marriage were rare, but it was common for the first birth to occur soon after marriage. Divorces were also rare: in 1970 the proportion divorced among ever-married men at age 50 was 4 per cent. Partnership and fertility behaviours changed rapidly during the 1970s and 1980s: by 1993, the proportion divorced had risen to 16 per cent, but for most birth cohorts included in our study sample, the societal context of family formation was traditional, with strong norms of marrying before childbirth and staying married (Finnäs 1995, 1996). While the Nordic countries are known for their family policies emphasizing gender equality, for our study cohorts parental leave existed essentially as maternity leave for women, although since 1980, fathers have been allowed to take one month of the mother’s quota for parental leave (Haas and Rostgaard 2011). Compared with most Western countries, women’s employment rates in Finland were already high soon after the Second World War, with the vast majority of women, including mothers of young children, working full-time (Rønsen and Sundström 2002). Nevertheless, men have continued to provide a larger share of household earnings (Nieuwenhuis et al. 2017), and fathers’ contribution to childcare and housework has increased very slowly, with women still bearing greater responsibility for unpaid work in families (Miettinen et al. 2015).

Data and methods

Data

Our study used a longitudinal 10 per cent sample drawn from the Finnish 1950 Household Census, which had been further linked to administrative register data on reproductive histories, to earnings information, and to hospital discharge and death records.

The data allowed us to identify brothers who lived, as children, in the same household in 1950 and to follow them until their early 50s. Data linkage was carried out by Statistics Finland using the personal identification codes assigned to all residents of Finland. We restricted the sample to men who were born in 1938–50, lived with their family at the time of the 1950 Census, and resided in Finland for at least three years between ages 48 and 52. Loss to follow-up due to emigration or death by age 48–52 was 19 per cent (n  =  10,363), resulting in our ‘all men’ sample (n  =  43,204). For our main analyses, we further excluded men without male siblings (49.2 per cent, n  =  21,243). This analytic sample of brothers consisted of 21,961 brothers in 9,443 families, among whom we examined the association between fatherhood timing and men’s earnings around ages 30, 40, and 50.

Fatherhood timing

Men’s age at first birth was based on month-of-birth information for the study individuals and their firstborn children in the Finnish population registers. We categorized fatherhood timing into ages <20, 20–24, 25–29, 30–34, and 35+. Men who had no biological children by age 52 were included in the analyses in a separate category, ‘no children’.

We knew that the register-based information on men’s reproductive histories was nearly complete for the cohorts studied, as a comparison between the cohort-specific lifetime fertility of men and women in the 1950 Census sample had indicated no notable bias attributable to unknown fatherhood (Einiö et al. 2016). However, stillborn babies and children who died before the introduction of personal identification codes in the late 1960s were not registered to their parents’ files. Fathering a child given up for adoption was not included in our measure of fatherhood due to absence of parental responsibilities. Becoming a father through adopting children was also excluded because the time of adoption was unknown.

Outcomes

Our main outcome for midlife earnings was the five-year average of annual wage and salary earnings measured at ages 48–52 (calendar years 1986–2002 depending on the cohort). We also used information on annual earnings within the age ranges of 28–32 and 38–42 (earlier data on earnings were available only for the quinquennial census years 1970, 1975, and 1980), to assess whether similar associations with fatherhood timing could already be observed at younger ages. Information on earnings originated from the Finnish Tax Administration. Earnings were adjusted for changes in consumer prices. The distribution of earnings may be skewed and prone to outliers. In the regression models, we used inverse hyperbolic sine (IHS)–transformed dependent variables to minimize the effect of extreme values on the results (Friedline et al. 2015). In addition, we ran a sensitivity analysis of our main results, excluding those with zero earnings.

Control variables

In the adjusted models, we controlled for the potential confounding effects of the men’s birth year, birth order, region of residence, attainment of at least secondary education, and hospital-treated health problems from early adulthood to age 35. Including such observed characteristics in the models is important, as the within-family approach controls for all factors shared within families but does not take into account heterogeneity between siblings (Holmlund 2005). In the between-family analyses on the samples of brothers and of all men, we also included maternal age at birth (which may affect children’s genetic endowment and also the environment they grow up in, thus indirectly influencing their midlife earnings (Barclay and Myrskylä 2016)) and total number of siblings (which may affect the availability of parental resources (Murphy and Wang 2001; Goodman et al. 2012; Kolk and Hällsten 2017)).

Variables for birth year (categorical), birth order (firstborn vs later-born siblings), number of siblings (1, 2, 3, 4+), and mother’s age at birth (<20, 20–24, 25–29, 30–34, 35+) were based on the 1950 Household Census. Firstborn children tend to achieve higher earnings in adulthood than younger siblings (Behrman and Taubman 1986), although the effects of birth order appear to be concentrated in earlier parts of the career (Bertoni and Brunello 2016).

Attainment of at least secondary education was used as a proxy for human capital investment and controlled for because of its links with both fertility timing and earnings (Card 1999; Nisén et al. 2014). To minimize issues of endogeneity in the relationship between educational attainment and fertility, we measured education by completion of secondary-level qualifications that are usually obtained below age 20. Region of residence (as measured by large (NUTS 2) areas in Finland, n  =  4) has been shown to be associated with reproductive behaviour (Kulu et al. 2007) and was also associated with earnings in our study sample. Education and region of residence were measured at ages 30–34 based on the 1970, 1975, and 1980 Censuses.

Severe health problems may affect both reproductive behaviour and earnings (Luft 1975; Einiö et al. 2016). Study individuals were categorized as having had either mental or somatic health problems if they had received inpatient hospital care for these causes between ages 18–28 and age 35 (the former depending on their birth cohort, as hospital records were available from 1969 onwards). The International Classification of Diseases (ICD-8) was used to separate mental health diagnoses from all other (somatic) diagnoses. We limited the measurement of severe health problems to early midlife, as our interest lies in the impact of health problems during the life period when most men become fathers for the first time. The distributions of the observed control variables in the two samples are displayed in Tables A1 and A2, supplementary material. We did not include marital status, because of the shared selection mechanisms into union formation and fatherhood (Trimarchi and Van Bavel 2017; Ludwig and Brüderl 2018; Mari 2019). Moreover, previous studies have shown no evidence of causal effects of marriage on men’s earnings (Loughran and Zissimopoulos 2009; Ludwig and Brüderl 2018), and the role of marriage in driving the effect of fatherhood is likely to be minor (Kunze 2020).

Methods

We used linear regression models with sibling fixed effects to assess the relationship between fatherhood timing and men’s midlife earnings among brothers (within-family differences). The sibling fixed-effects models allowed us to control for the unobserved social and genetic confounders common to same-sex siblings that may otherwise have biased the results (Allison 2009). To account for sibling heterogeneity, the observed control variables that may differ between siblings (birth year, birth order, region of residence, education, and health problems) were added to the within-family models as categorical variables. For comparative purposes, we also estimated ordinary least squares regression models to assess the relationship between fatherhood timing and midlife earnings, controlling only for observed confounders (between-family differences). These models involved the same study sample of brothers and included the same observed control variables as the sibling fixed-effects models plus maternal age at birth and number of siblings. As a sensitivity analysis, we also ran the between-family models for the sample of all men, which includes men without siblings and those with female siblings only. The standard errors were clustered by family in all the models. We present adjusted and unadjusted results for both the between- and within-family models.

Finally, we used unconditional quantile regression (UQR) models to assess the influence of fatherhood timing on the overall distribution of fathers’ earnings (Firpo et al. 2009; Rios-Avila 2020). As the estimates from the UQR models capture the combined effects of the distribution of fatherhood timing, the distribution of earnings in the study sample, and the individual-level association between fatherhood timing and earnings, they cannot inform us about individual-level effects (Borgen et al. 2023). Instead, they allow us to assess the population-level influence of fatherhood timing. The UQR coefficients inform us how a hypothetical change in the share of men in the respective category of fatherhood timing (e.g. the share of young fathers) would alter the quantile values of the fathers’ earnings distribution. For ease of interpretation, we present the results as marginal unit changes from a hypothetical five-percentage-point increase in the share of men in the respective category of fatherhood timing (Rios-Avila 2020; Rios-Avila and de New 2022). We estimated the UQR models on the sample of all men (n  =  43,204), both excluding and including those with no children (n  =  8,243, 19 per cent), adjusting all models for the observed control variables. As a sensitivity analysis, we replicated the same analysis for the sample of all men but excluding those with zero midlife earnings (n  =  3,335, 8 per cent).

Results

Table 1 shows the descriptive statistics on annual earnings by fatherhood timing for the sample of brothers. Men who became fathers at earlier ages achieved lower earnings at ages 48–52 and belonged to the lowest earnings decile more often than men who sired their offspring at later ages. For example, 10 per cent of ‘adolescent’ fathers (aged <20) and 8 per cent of ‘young’ fathers (aged 20–24) belonged to the lowest earnings deciles, as opposed to around 6 per cent of men who became fathers at older ages. In contrast, the share of men in the highest earnings decile was clearly largest among men who became fathers at ages 25–29. The differences in earnings were similar but less pronounced around age 40, while earnings around age 30 tended instead to be higher among men who were already fathers by that age. The lowest earnings were consistently found among men without children, and this group was highly over-represented (23 per cent) in the lowest midlife earnings decile. Differences across the earnings distribution are further illustrated in Figure 1, which shows kernel density plots of midlife and earlier earnings by fatherhood timing.

Figure 1 Distributions and kernel density plots of annual earnings in euros at (a) ages 48–52; (b) ages 38–42; and (c) ages 28–32 by fatherhood timing, relative to first-time fathers aged 25–29: all men sample, Finland

Note: Highest earnings capped at 100,000 euros; k refers to 1,000.

Source: Authors’ analysis based on Finnish 1950 Census 10 per cent sample linked to administrative register data (Statistics Finland and Finnish Institute for Health and Welfare).

Table 1 Descriptive statistics of men’s annual earnings at ages 48–52, 38–42, and 28–32 by fatherhood timing: study sample of brothers, Finland

	Age at fatherhood
	All	<20	20–24	25–29	30–34	35+	No children
Annual earnings (euros) at ages 48–52
Mean (SD)	27,801 (45,888)	24,023 (18,015)	27,683 (23,149)	33,655 (76,385)	30,650 (24,126)	29,081 (21,960)	17,620 (18,705)
Median (IQR)	26,638 (29,103)	25,094 (28,057)	27,499 (26,967)	30,184 (28,236)	28,855 (26,770)	28,140 (24,773)	12,721 (28,737)
Share in lowest decile (percentage)	10.0	10.1	8.2	5.9	5.5	6.3	22.8
Share in highest decile (percentage)	10.0	5.6	8.7	14.6	12.3	10.3	4.4
Annual earnings (euros) at ages 38–42
Mean (SD)	28,343 (18,334)	26,840 (14,848)	29,397 (17,277)	32,188 (19,047)	30,348 (19,742)	28,140 (16,654)	19,794 (16,132)
Median (IQR)	27,114 (16,388)	27,407 (14,021)	28,111 (14,443)	29,323 (16,997)	28,204 (15,558)	26,130 (16,314)	20,431 (23,099)
Share in lowest decile (percentage)	9.8	9.9	7.0	5.4	6.7	7.6	23.4
Share in highest decile (percentage)	9.8	6.3	9.1	14.0	10.9	9.4	4.5
Annual earnings (euros) at ages 28–32
Mean (SD)	23,477 (15,231)	24,922 (12,471)	25,265 (15,861)	26,064 (14,466)	23,622 (17,080)	21,671 (12,679)	16,654 (12,988)
Median (IQR)	22,724 (13,476)	24,466 (11,476)	23,989 (11,936)	24,413 (13,702)	22,517 (13,734)	21,564 (15,250)	17,245 (17,920)
Share in lowest decile (percentage)	9.7	6.0	5.3	5.8	9.6	11.9	23.0
Share in highest decile (percentage)	9.7	10.7	9.8	13.6	10.2	8.0	3.7
N (percentage)	21,961 (100)	587 (2.7)	7,054 (32.1)	6,541 (29.8)	2,336 (10.6)	1,180 (5.4)	4,263 (19.4)

Note: SD refers to the standard deviation; IQR refers to the interquartile range.

Source: Authors’ analysis based on Finnish 1950 Census 10 per cent sample linked to administrative register data (Statistics Finland and Finnish Institute for Health and Welfare).

Becoming a father at adolescent ages was relatively rare (2.7 per cent in the sample), but young fathers (aged 20–24) represented nearly one-third of the study sample (Table 1). Adolescent and young fathers were more likely than others to have been born to younger mothers and to be firstborn themselves (distribution of the men’s socio-demographic characteristics by fatherhood timing shown in Tables A1 and A2, supplementary material). Compared with men who became fathers at ages 25–29, younger fathers were also less likely to have attained at least secondary education and more likely to have experienced somatic and mental health problems.

Midlife earnings (around age 50) remained lower among adolescent and young fathers than among men who became fathers in their late 20s, when controlling for observed differences in their education, health, and demographic characteristics (Figure 2(a), unadjusted and adjusted between-family models). For ease of interpretation, Table 2 presents the regression coefficients from the adjusted between-family models with the effect size quantified in euros. For example, adolescent fathers earned on average 2,842 euros (95 per cent CI 628–4,854) less and young fathers 1,869 euros (874–2,573) less in midlife than those who became fathers at ages 25–29. The economic disadvantage among adolescent and young fathers was also observed in terms of their earlier earnings around age 40 (Figure 2(b)) but not around age 30 (Figure 2(c)).

Figure 2 The association between fatherhood timing and annual earnings at (a) ages 48–52; (b) ages 38–42; and (c) ages 28–32: between- and within-family models using all men sample, Finland

Notes: IHS-transformed coefficients with 95 per cent confidence intervals, with first-time fatherhood at ages 25–29 as reference. Adjusted models include birth year, birth order, region of residence, education, and health problems, plus maternal age and number of siblings in between-family models.

Source: As for Figure 1.

Table 2 Regression coefficients from adjusted between-family models for the association between fatherhood timing and annual earnings at (a) ages 48–52; (b) ages 38–42; and (c) ages 28–32: sample of brothers, Finland

(a) Annual earnings at ages 48–52
Timing	b	Lower CI	Upper CI	Euros	Lower CI	Upper CI
<20	−0.359	−0.613	−0.105	−2,842	−4,854	−628
20–24	−0.236	−0.325	−0.146	−1,869	−2,573	−874
25–29	(ref)
30–34	0.011	−0.103	0.125	87	−816	748
35+	−0.047	−0.201	0.107	−372	−1,591	640
No children	−1.728	−1.862	−1.593	−13,682	−14,743	−9,533
(b) Annual earnings at ages 38–42
Timing	b	Lower CI	Upper CI	Euros	Lower CI	Upper CI
<20	−0.305	−0.523	−0.086	−4,021	−6,896	−1,134
20–24	−0.082	−0.151	−0.013	−1,081	−1,991	−171
25–29	(ref)
30–34	−0.109	−0.204	−0.014	−1,437	−2,690	−185
35+	−0.163	−0.300	−0.026	−2,149	−3,956	−343
No children	−1.338	−1.455	−1.221	−17,642	−19,185	−16,099
(c) Annual earnings at ages 28–32
Timing	b	Lower CI	Upper CI	Euros	Lower CI	Upper CI
<20	−0.009	−0.197	0.179	−104	−2,281	2,072
20–24	0.088	0.021	0.156	1,019	243	1,806
25–29	(ref)
30–34	−0.281	−0.389	−0.173	−3,253	−4,504	−2,003
35+	−0.359	−0.498	−0.219	−4,156	−5,766	−2,536
No children	−1.258	−1.373	−1.143	−14,565	−15,896	−13,234

Notes: The table shows results from adjusted between-family models among the study sample of brothers (n  =  21,961), controlled for birth year, birth order, region of residence, education, and health problems. Coefficients (b) and their 95 per cent confidence intervals (CI) are presented both with IHS transformation and as transformed into euros. The coefficients are converted into euros with ½ × (exp(y) + exp(−y)) × b (Pence 2006, p. 6), y = mean of the IHS-transformed earning, and b = coefficient. Ref is the reference category.

Source: As for Table 1.

Men who became fathers at ages 30 or later displayed very similar midlife earnings to those aged 25–29 (Figure 2(a) unadjusted and adjusted between-family models). However, earlier earnings, particularly around age 30, were somewhat lower among men who became fathers at ages 30+ than among the younger fathers (Figure 2(c), Table 2(c)). While the differences were attenuated slightly in the adjusted models, the lowest earnings were clearly and consistently observed among men with no children.

The lower midlife earnings of adolescent and young fathers were to a further degree attributable to unobserved characteristics shared by siblings, as well as to observed differences in their socio-demographic and health characteristics (Figure 2(a), within-family models). A similar pattern of attenuating differences was also observed around age 40 (Figure 2(b)). In contrast, among men who became fathers at later ages, adjustment for observed confounders or within-family comparison made little difference to the associations (which were modest to begin with). Overall, with the exception of men with no children, all the associations between fatherhood timing and midlife earnings became negligible in a within-family comparison.

To assess the population-level influence of fatherhood timing on the earnings distribution of men, we estimated UQR coefficients among both the study sample of all men and a subsample of fathers only. The coefficients in Figure 3 show how a marginal unit change, in this case a hypothetical five-percentage-point increase in the share of men in the respective category of fatherhood timing, would have altered the quantile values of the earnings distribution (coefficients and their 95 per cent confidence intervals for the models are shown in Tables A4 and A5, supplementary material). For the fathers subsample, an increase in the share of adolescent or young fathers would have shifted the midlife earnings distribution to the left, lowering the quantile values somewhat more at the highest earnings deciles, as well as in the deciles below the median (Figure 3(a)). These changes in the overall earnings distribution were rather similar in these two age groups, despite their clear difference in size. As expected, an increase in the share of men who had become fathers at ages 25–29, often the highest earners, would have had the opposite effect, inducing a distributional shift to higher earnings, while increases in late fatherhood would have made little difference to fathers’ earnings distribution. Overall, the very lowest end of the fathers’ midlife earnings distribution would have been largely unaffected by changes in the relative shares of fathers of different ages.

Figure 3 Unconditional quantile regression coefficients of fatherhood timing on the earnings distributions at (a) ages 48–52; (b) ages 38–42; and (c) ages 28–32: fathers and all men, Finland

Notes: The coefficients with 95 per cent confidence intervals (shaded areas) show the marginal unit change in quantile values of the earnings distribution following a hypothetical five-percentage-point increase in the share of men in the respective category of fatherhood timing. All models are adjusted for birth year, birth order, maternal age, number of siblings, region of residence, education, and health problems.

Source: As for Figure 1.

In the models including all men, the negative distributional shifts relating to early fatherhood were smaller, paling in contrast to those induced by the hypothetical increase in the share of men with no children (Figure 3). Due to the potential instability of the UQR estimates because of a high proportion of zero earnings in the lowest decile, we ran a sensitivity analysis, excluding all men with zero earnings. The distributional shifts were somewhat smaller in these models—in particular, for changes induced by the share of childless men to the quantile values below the median—but the overall patterns remained very similar (Figure A1, supplementary material).

The influence of fatherhood timing on the distribution of earlier earnings (around age 40) was weaker but otherwise mostly similar to midlife earnings, whereas around age 30, increases in early fatherhood would have induced modest distributional shifts to higher earnings and increases in late fatherhood shifts to lower earnings (Figure 3(b)–(c)).

Discussion

Based on longitudinal census data with over 50 years of follow-up, this study used a within-family approach to investigate the associations between fatherhood timing and men’s midlife earnings. Relative to men who had their first child at ages 25–29, adolescent fathers (aged <20) and young fathers (aged 20–24) experienced lower earnings around age 50, but these associations became negligible after accounting for unobserved characteristics shared by brothers and observed sibling differences in education, health, and demographic characteristics. Later-than-average fatherhood was not associated with midlife earnings, while having no children was associated with a substantial earnings disadvantage. At the population level, increases in the share of adolescent or young fathers would have induced clear negative distributional shifts in fathers’ midlife earnings. However, among all men, any influence of fatherhood timing on the midlife earnings distribution paled in comparison with that of childless men.

Previously, Dariotis et al. (2011) and Assini-Meytin and Green (2015) reported subsequent economic disadvantage among adolescent and young fathers in the US. Dariotis et al. (2011) also indicated that the relative disadvantage associated with early fatherhood increased over the life course. Our results corroborate this finding and with a substantially longer follow-up. The slightly more pronounced differences in earnings around age 50 than age 40 and the absence of any early fatherhood penalty in earnings around age 30 are consistent with the interpretation that young fathers get a good start, but due to early parental responsibilities hampering investments in education and training (Sigle-Rushton 2005), their earnings stagnate in the long term. That the largest differences were at the lowest end of the midlife earnings distribution (Figure 1(a)) also highlights young fathers’ higher likelihood of less stable employment. Nevertheless, our results comparing brothers suggest that the midlife earnings disadvantage of adolescent and young fathers is largely explained by selection into early fatherhood by unobserved social and genetic characteristics: a finding that is in line with recent longitudinal studies emphasizing the role of selection behind any fatherhood premium in wages (Mari 2019; Kunze 2020; Icardi et al. 2022).

Men who became fathers in their late 20s or older all achieved very similar midlife earnings. Earlier earnings, around age 30, were, however, lower among men who had no children yet but would later become fathers (even among brothers); this could lend some support to the hypothesis of employer discrimination against non-fathers. The lower earlier earnings in this group could, however, also be explained by delayed employment careers due to pursuing higher education. By their early 50s, and to a large extent by their early 40s, the later-than-average fathers had already caught up with the earnings of the reference group.

The lowest earnings were consistently found among men who had fathered no children by their early 50s, and a clear difference between fathers and non-fathers persisted even when comparing brothers. The proportion of men with zero or very low midlife earnings was also substantial among those with no children, suggesting high rates of long-term unemployment and/or work disability. Economic disadvantage relating to weak attachment to the labour market has been ignored in studies restricted to those in steady employment, and this could partly explain the disparity between our results and recent within-individual or within-family studies that have found no notable fatherhood premiums among employed men (Mari 2019; Kunze 2020; Icardi et al. 2022). However, in a sensitivity analysis of our main results that excluded those with zero earnings, the observed patterns remained, although the differences in earnings were overall much smaller (Figure A2, supplementary material).

There are several possible explanations for the less advantaged economic situation of non-fathers. First, these men may have had difficulties in starting a family due to poorer health. Our analysis provided some support for the health selection hypothesis, as health problems by age 35 were more prevalent among men with no children (Table A1, supplementary material) and after adding controls for mental and somatic health problems, the differences were attenuated (results not shown). Second, men without children were also likely to be negatively selected according to their economic prospects, as their earnings around age 30 were already substantially lower than those of other men, with almost one in four belonging to the lowest earnings decile. Finally, men who had fathered no children were likely to be selected by other (unmeasured) characteristics associated with reproductive behaviour: for example, individual fertility preferences, personality traits, or physical attractiveness (Jokela 2009; Jokela et al. 2011; Balbo et al. 2013).

Our analyses using UQR models suggested that distributional shifts induced by changes in the share of adolescent or young fathers, or the share of childless men, tended to be stronger at the higher end of the earnings distribution, while the very lowest end remained largely unaffected. These population-level estimates capture the distributional effects of both earnings and fatherhood timing, as well as the individual-level associations between the two (Rios-Avila and de New 2022; Borgen et al. 2023). Although belonging to the lowest earnings decile was common (23 per cent) among childless men, and more common among adolescent and young fathers than other fathers, it is likely that childbearing patterns account for a relatively modest part of the overall risk of very low earnings in midlife. This is in line with our within-family analyses and emphasizes the important role of selection behind the earnings disadvantage of adolescent and young fathers and childless men.

Strengths and limitations

This study was based on longitudinal register data with a modest degree of attrition due to death or emigration and a lack of problems with non-response or recall bias regarding earnings, reproductive histories, family social background, or individuals’ socio-demographic and health characteristics. Moreover, the large sample and the unique linkage of siblings who lived in the same household in 1950 allowed us to compare earnings between brothers.

The main strength of the within-family approach is that it reduces the selection bias related to the unobserved social and genetic characteristics shared by siblings. Although not as similar as identical twins, brothers share more of their social networks, family resources, and genetic characteristics than do randomly selected men. Such characteristics are often difficult or impossible to measure, and the within-family comparison provides an opportunity to control for them to a degree without direct measurements.

The within-family approach has its limitations. By design, the study sample consisted of siblings and was thus not representative of the total population. However, having siblings was common among the cohorts included in this study: 55 per cent of men in the original sample had a brother born within the same period (1938–50). Compared with all men, average earnings were slightly lower among the sample of brothers, who were also slightly less educated (for distributions among all men, see Tables A2 and A3, supplementary material), possibly reflecting dilution of parental resources in larger families (Kolk and Hällsten 2017). Overall, however, the results from sensitivity analyses on the sample of all men were highly similar to the between-family results based on the sample of brothers (model estimates for both samples are shown in Figure A3, supplementary material).

Although brothers are much more similar to each other than individuals selected randomly from a population, brothers who became fathers at a young age or who delayed fatherhood until later ages may still differ by unobserved characteristics that affected both fatherhood timing and later earnings. For example, personality factors, such as emotional stability, autonomy, extraversion, agreeableness, and conscientiousness could affect union formation and childbearing, as well as earnings. It has been suggested that emotional stability is associated with higher earnings among both men and women (Nyhus and Pons 2005) and, presumably, emotional stability could also be more prevalent among brothers who fathered their first child at the more normative childbearing ages. Nevertheless, the impact of personality traits on the timing of childbearing has been shown to be minor (Jokela et al. 2011).

Previously, teenage motherhood has been shown to have negative effects that extend to the outcomes of siblings, thus understating the true effect of early parenthood when comparing siblings (Heissel 2021). However, while there is currently no evidence of such spillover effects relating to early fatherhood, it is possible that early parenthood for one sibling might affect other siblings’ outcomes due to reallocation of resources within families, for example (Geronimus and Korenman 1993).

In this study, we controlled for observed confounders, including men’s year of birth, own birth order, education, region of residence, and mental and somatic health problems. However, the role of these factors, education in particular, in the associations between fatherhood timing and later earnings may be complex (Nisén et al. 2022). On the one hand, while lower education tends to be associated with earlier childbearing (Nisén et al. 2014), early fatherhood could also lead to a decision to invest in parenting and providing for the family instead of in personal education. In this case, education would be a mediator rather than a confounder in the association between fatherhood timing and midlife earnings. On the other hand, higher education is correlated with higher earnings, and pursuing higher education may lead to postponement of family formation and childbearing (Nisén et al. 2014). For these reasons, we included in our analyses only the attainment of secondary qualifications that are usually completed below age 20 and so could not consider the full complexity of the associations between education and fertility. Overall, while we could not observe whether becoming a father at a certain age was down to choice or chance for these men, the observed background variables and unobserved family background factors are likely to have captured at least some degree of their differing levels of childbearing intentions.

The relevance of our findings for different societal contexts and younger cohorts remains an open question. The Finnish cohorts included in this study were born in 1938–50, and nearly all these men fathered their children in the traditional context of marriage in a society with low rates of divorce (Finnäs 1995, 1996). Among more recent birth cohorts, the availability of and advances in birth control and other reproductive health products and services have allowed for more individual control over reproductive choices. Coupled with increasingly complex partnership behaviours and continued postponement of childbearing, these could be increasingly important mechanisms behind any associations between fatherhood timing and later earnings among more recent cohorts. For example, a recent study from the US suggested that the fatherhood premium in wages depends on whether the father is co-resident with the child or not (Gowen 2023). The increasing participation of fathers in family life and sharing of parental responsibilities may also render men more vulnerable to the economic consequences of career disruptions. Interestingly, in contrast to most recent studies that have found either modest or non-existent fatherhood premiums in earnings, a Norwegian register study (Cools and Strøm 2016) found a small fatherhood wage penalty that was explained largely by paternity leave. The effects were stronger among men who had their first child in 1993 or later, consistent with the increasing participation of men in child-rearing (Cools and Strøm 2016). Also in Finland, taking paternity leave in the 2000s has been found to be associated with modest decreases in wage trajectories (Morosow and Cooke 2022).

Conclusions

In this longitudinal census-based study, we observed a long-term earnings disadvantage among adolescent and young fathers, as well as for men with no children. However, when limiting the comparison to brothers, the associations between early fatherhood and later earnings largely disappeared. Although our findings emphasize the role of selection into early fatherhood, the associations and the factors behind them could differ among more recent cohorts. We encourage future research to delve deeper into the complex mechanisms through which early parenthood may contribute to the high probability of young fathers being among those with the very lowest midlife earnings.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 All authors are based in the Helsinki Institute for Demography and Population Health, University of Helsinki. Please direct all correspondence to Hanna Remes, Helsinki Institute for Demography and Population Health, University of Helsinki, P.O. Box 42, FIN-00014 Helsinki, Finland; or by E-mail: [email protected].

2 The permissions to use the register-based data were obtained from Statistics Finland and the Finnish Institute for Health and Welfare (permissions TK-53-789-10 and THL/1273/6.02.00/2013). The register data were collected originally for administrative and statistical purposes, and the use of such data for purposes of scientific research carried out in public interest does not require informed consent from participants.

3 Funding: Pekka Martikainen was supported by the Academy of Finland (grant agreement 1308247).