4 Earnings impacts of childbirth: Empirical framework

View the most recent version.

4.1 Previous researches

The earnings differences between mothers and other women, also sometimes referred to as the family gap, are well documented in many developed countries. On average, the raw earnings differences were around 20% in the United States and the United Kingdom; in Canada, it was estimated to be 17%.¹⁷ A number of hypotheses have been proposed by researchers.

First, some authors attribute the family gap to unmeasured heterogeneity. The differences could be spurious if motherhood is negatively correlated with certain unobserved characteristics while the pay rate is positively correlated with these characteristics. An early study by Korenman and Neumark (1992) and a more recent study by Gupta and Smith (2002) show that the family gap between mothers and non-mothers decreased substantially, or disappeared completely, once individual fixed effects were controlled for.

Second, a related hypothesis is that the choice to become a mother is itself endogenous. Studies on female labour supply generally find that fertility is negatively correlated with the pay rate.¹⁸ In the current context, since low pay implies the opportunity cost for a woman to separate from her job for family responsibilities is low, one or more years of low earnings may cause a woman to become a mother. A number of studies—such as Gronau (1988), Korenman and Neumark (1992), Edin and Nynabb (1992), and Belzil and Hergel (1999)—find no strong support for the hypothesis that low pay causes women to take time out for childbearing and rearing. But other studies—for example, Lundberg and Rose (2000)—find evidence of negative selection into parenthood, while Viitanen (2004) concludes that endogenous participation and fertility account for about 50% of the total family gap.

Third, the human capital theory generates a few explanations for the family gap. One of the hypotheses is based on the fact that most women interrupt their careers for childbearing and rearing and, as a result, accumulate fewer years of working experience. Also, during extended interruptions, mothers may also experience human capital depreciation. Most studies suggest that human capital losses are important factors in explaining the family gap, but these losses alone cannot explain all of the wage or earnings penalties. For example, Phipps, Buirton and Lethbridge (2001) find that controlling for experience accumulation and human capital depreciation reduces the earnings penalty from 17% to 12% in Canada, while Waldfogel (1998b) concludes that around 30% to 40% of the pay penalty can be explained by mothers' lower levels of experience and lower returns to their experiences.

Fourth, the work effort hypothesis asserts that child rearing and caring often create family and work conflicts for mothers and that these conflicts reduce the work effort of mothers. For example, Becker (1991) predicts that the extra household tasks associated with child rearing and caring consume a lot of energy from the mothers of young children and make them less productive at work than childless women. By including hours spent on unpaid household tasks in their analysis, Phipps, Buirton and Lethbridge (2001) find that the earnings penalty fell from around 13% to 7%. On the other hand, Anderson, Binder and Krause (2003) argue that since high-skilled jobs typically require more effort, the wage penalty would rise with the mothers' years of education. Their finding that the medium-skilled (high-school graduate) mothers suffer more prolonged and severe wage penalties than do either low- or high-skilled mothers casts some doubt on the work-effort explanation.

Related to the work-effort hypothesis is the self-selection hypothesis that predicts that potential mothers may choose 'mother friendly' jobs that have higher initial earnings but flatter the earnings growth path.¹⁹ Budig and England (2001) examine whether mothers choose less energy-demanding occupations. They conclude that mother-friendly job characteristics explain little of the wage penalty incurred by mothers. Waldfogel and Mayer (2000) also find that controlling for occupation does not eliminate motherhood wage penalties.

Finally, some authors interpret the earnings differences between mothers and other women as a result of statistical discrimination, under which mothers earn less than childless women because employers expect them to be less productive because of child-rearing responsibilities. For example, Waldfogel (1995) attributes the unexplained portion of the family gap—after controlling for human capital and other variables—to employer discrimination or other causes.

While the aforementioned studies have improved our understanding toward the earnings gaps between mothers and other women, they are subject to a number of limitations. First, the vast majority of previous studies have failed to examine how the earnings differences had changed over the post-childbirth years. We contribute to the literature by investigating the earnings of mothers over a long period of time, including the years before childbirth. By examining how the earnings of a mother change before she gives birth, we are able to see if declining earnings cause a woman to self-select into motherhood.²⁰ By examining the earnings dynamics over many post-childbirth years, we are able to see if the family gap is permanent or temporary.

Second, our study is based on a very large administrative dataset. This not only allows us to construct a sample of women who became mothers over a number of years, it also permits us to construct a sample of women who did not give birth. With a sample of 'non-mothers,' we are able to borrow, from the program evaluation literature, statistical techniques that help us to obtain more reliable estimates of the earnings differentials.

Finally, unlike other studies, we also examine the effect of maternity benefits on the earnings changes of the mothers. Since mothers who temporarily withdraw from their jobs for maternity purposes typically receive certain amounts of compensation from the government—for example, in Canada, women who held paid jobs for a certain period of time were eligible for the Employment Insurance benefits during their maternity/parental leaves—ignoring the maternity benefits may lead one to overestimate the motherhood earnings penalties.

4.2 Empirical model

To model the earnings trajectories of mothers, and to take advantage of the strength of our data, we adopt a model developed by Jacobson, LaLonde and Sullivan (1993) in analysing the earnings losses of displaced workers in the U.S. state of Pennsylvania.²¹ The model is characterized by a very flexible age–earnings profile that allows workers to have different earnings intercepts and earnings growth paths.

where y_it denotes the (log) annual earnings of woman i in year t, X_it is a vector of observable worker or job characteristics, α_i are individual-specific fixed effects, while ω_it are the worker- specific earnings growth paths. ε_it is the idiosyncratic error term and D_it^k is a vector of dummy variables that equal 1 if worker i gives birth k years prior to year t (and D_it^k = 0 otherwise). Notice that the worker-specific earnings growth path, ω_it, allows each woman to have her own person- specific slope of the earnings profile.

In order to examine the earnings changes of a mother around the years she gives birth, we specify a = -3 and b = 9.²² The choice that a = -3 allows us to investigate how the earnings of a mother would change during the three-year period prior to childbirth. The estimated earnings changes over these years can be used to test if declining earnings cause women to leave their jobs for childbearing and rearing (the endogenous motherhood hypothesis). With b = 9, we can examine the earnings effects of childbirth over nine post-childbirth years.

When the restrictions that α_i = 0 and ω_i = 0 are imposed, the general model becomes a simple regression model, while if the restriction ω_i = 0 alone is imposed, we obtain the fixed-effects model. In the case where the earnings differences between mothers and other women are purely due to unobserved heterogeneity, the fixed-effects estimates of earnings effects of childbirth would be equal to 0. Hence the fixed-effects model allows us to test if earnings differences between mothers and other women are spurious.

The fixed-effects model also helps to reduce the potential bias caused by omitted variables that are constant over time—immigration status, for example. However, estimates from the fixed-effects model can be biased since, on the one hand, not all omitted variables are constant over time; for example, education and experience. On the other hand, mothers may self-select into jobs with flatter earnings-growth paths. Hence, the model that allows all women to have different slopes of the earnings profiles is more general and flexible than the fixed-effects model.

 

¹⁷ The family gap is often measured in the hourly wage rate. For Canada, this was measured by personal income for women aged from 25 to 54 who worked full time during 1995. See Phipps, Buirton and Lethbridge (2001).

¹⁸ See Browning (1992) for a survey.

¹⁹ The slower earnings growth is also consistent with low (initial and subsequent) human capital investment of mothers who anticipate career interruptions.

²⁰ We examine the earnings of the potential mothers for up to three years before the child is born. We notice that lower earnings related to sickness due to pregnancy might occur during the months before the child is born, and these earnings dips have nothing to do with the self-selection issue.

²¹ Ejrnaes and Kunze (2004) is the only study that follows this approach. The current work is different from theirs in two aspects: we allow different workers to have different earnings growth paths; and, we allow women to give more than one birth (see Section 6).

²² The conclusion by Browning (1992) that children aged up to 10 years could be as costly as infants partially motivated our choice of b=9. The choices for a and b involve a trade-off between more cohorts of mothers with estimates of earnings losses over fewer years around childbirth and fewer cohorts of mothers with estimates of earnings losses over more years around childbirth.