13(38)
4
Data and methodology
For the empirical analysis register data from the Swedish Social Insurance
Agency are used. Data are assembled from records obtained from local
insurance offices and cover the entire Swedish population. They contain
detailed information on the starting date of parental leave, the number of
days (parts of the day if not a full day) and the amount of benefit per day.
They also include parents’ individual characteristics such as gender, date of
birth, birth order of the child, geographical location, earnings, educational
level and country of birth.
Empirically, we make use of the fact that all three reforms were introduced
for children born after a specific date. The first reserved month is applicable
to children born on or after January 1, 1995, the second reserved month to
children born on or after January 1, 2002, and the gender equality bonus to
children born on or after July 1, 2008. Thus, children born within a few
days of each other are treated under different regulations, which makes all
three reforms examples of natural experiments (see Rosenzweig and
Wolpin 2000, Angrist and Krueger 2000).
The birth of a child may be seen as a random event, both as the time of
conception cannot be completely controlled by the parents, but also as the
duration of pregnancy is normally distributed with a mean of 40 weeks with
a standard deviation of two weeks. The birth of a child cannot be
postponed, and triggering a birth (except for health reasons) is considered
unethical and against professional standards for medical personnel (Ekberg
et al. 1995).
From the register data all parents with children born from two weeks before
to up to two weeks after each reform are sampled and subsets of parents
of children born before (control group) and after (treatment group) each of
the reforms are constructed. To control for potential seasonal variation in
the use of parental leave, we use a difference-in-difference approach
including parents with children born one year before the introduction of
each of the reforms in the statistical analysis.
As the reforms only affect children of parents with joint custody, children
whose parents do not have joint custody during the entire period of
observation are excluded from the sample. Multiple births, foreign-born and
adopted children are also excluded as there are special regulations for
these children. As the focus in this paper is on the division of parental leave
between women and men, children whose parents are of the same sex are
also excluded. Finally, children who died or emigrated during the period of
study are excluded from the sample. The final samples consist of 11 000 to
15 000 children for each reform studied.
The reliability of the empirical results crucially depends on two
assumptions. The first is that no other change that affects treatment and
control groups differently occurs at the same time as the reforms. The
second assumption is that there is no endogenous sorting at the time of the
reforms, that is, that some parents postpone or bring forward conception,
and thus the birth of the child, as a consequence of the information that the
reform will take place. If these assumptions do not hold, estimates may be
biased.
