20(38)
Also, the second reserved month shifted fathers’ distribution of used
parental leave days to the right, with a new peak at around 60 days (seven
to eight weeks). The distribution is less centered and there is more
variation in use after the second reserved month was introduced. For
mothers, the group using the maximum number of days increased in the
treatment group.
5.2
Logit models
To test the potential effect of the three reforms further we now proceed to
present results from logit models, where the dependent variable express
fathers’ risk of using more than a certain number of parental leave days.
9
The dependent variable is categorized in sixteen different dichotomies to
test whether the reforms impact at specific levels of usage.
For each reform, the independent variables and samples are identical in all
models. The dependent variables are the risk that the fathers are using
more than zero earnings-related parental leave days, using more than one
week of earnings-related parental leave days, using more than two weeks
of earnings-related parental leave days, and so on, up to the risk that the
fathers are using more than thirteen weeks of earnings-related parental
leave days. The models are specified as:
Logit(Y
1/0
| X, γ, δ, λ) = α +
X
i
’
β + γ Year + δ Month + λ Treatment
where
X
i
is a vector of individual characteristics of the father and the
mother, including age, country of birth, the birth order of the child, and
education (see Appendix E for details). The variables “Year” and “Month”
are dummy variables indicating the year and month of birth of the child,
10
where “Year” is one for children born around the introduction of each
reform and zero for children born a year earlier. The variable “Month”
assumes the value one for the month where each of the reforms were
introduced and zero for the month before the reform. The potential impact
of the reforms is measured by the variable “Treatment”, an interaction
variable of Year*Month, indicating that the father belongs to the treatment
group. It is the estimated value of this variable for each of the models that
is presented in Table 4.
11
For sensitivity analyses, we conducted logit models with only the
Treatment-dummy included, logit models with children born one week and
three days around the reform cut-offs respectively, as well as separate
analyses for various subgroups. In neither of the cases did the results differ
significantly from the main results.
Table 3 presents the results of the logit models for the “Treatment”
variable, with results expressed in odds ratios. In the first column of the
table the results for the first reserved month are presented. The results
show that there is a clear, significant effect of the first reform. The risk of
9
We have also conducted the same analysis where the dependent variable in the
corresponding models is the risk of using a certain number of parental leave days
(that is, being in the range of 0, 0.1-1 week, 1.1-2 weeks, …, more than 13 weeks,
compared with not being in that range). The results from these models are
presented in Appendix C. The results tend in the same direction as the main results
and strengthen the conclusions of the study.
10
Remember that to control for potential seasonal variation in the use of parental
leave, children born one year earlier than the control and treatment groups for
each reform are included in the logit-models.
11
Examples of the full model are presented in Appendix D.
