WP 2014-2

14(20)

parents, their partners for example, about the monitoring and this might

affect the benefit claims of the comparison group.

If such spill-over

effects exist, the estimated effect is biased downwards.

4.1

Results

Figure 2 shows the estimated differences in benefit days between the

treatment and the comparison group each week from 52 weeks before the

treatment assignment to 52 weeks after the treatment assignment. The

estimated difference between the treatment and the comparison group is

zero and statistically insignificant before the monitoring assignment takes

place. After monitoring the benefit take-up decreases in the treatment

group compared to the comparison group. The figure shows that the

monitoring effect appears after two weeks, which is expected since the

actual monitoring takes place when the parent applies for the benefit, not

when the parent is selected for monitoring.

Monitoring decreases the

benefit take-up by roughly 2 per cent during the first two months. The

figure shows that the effect of monitoring decreases over time and the

effect on benefit days is statistically significant up to roughly four months

after the monitoring.

In order to summarize the results as well as to study effect heterogeneity

we also estimate yearly effects using a difference in difference framework.

Here the outcome is total temporary parental benefits one year before and

one year after being assigned to treatment or control a given week. In the

estimation we also add control variables. The yearly estimates are given in

table 2.

From this table we can see that monitoring decreases the use of the benefit

by about 0.13 gross benefit days, which corresponds to approximately 1.4

per cent decrease in the benefit days. Similar results are also achieved

when measuring the benefit take-up as net benefit days. The results are

not sensitive to excluding parents with extremely few or many benefit days

before the monitoring. Adding controls for parent’s individual characteristics

or excluding the fixed effects for the monitoring week does not alter the

results (see row one and two in Table 2).

Effects for different subgroups are estimated with less precision. There are

no differences between men and women. We can however see that parents

with lower education react stronger to monitoring than parents with a

higher level of education.

This means that the treatment also affect the non-treated parents. This implies

that the no interference assumption in the stable unit treatment value assumption,

SUTVA, is not valid (see Rubin 1978). Empirical evidence suggests that such spill-

over effects exist when individuals are exposed to treatment in the sickness

insurance (see e.g. Hesselius et al. 2013).

The monitoring assignment is done on the first day of absence when the parent

notifies the SSIA. The benefit application can be sent in later. Between October

2010 and December 2013 the parents applied for the benefit on average 17 days

after reporting the child’s illness to the SSIA.

It seems thus, as the monitoring does not have a permanent effect on the

behavior of the monitored parents. However, one should note that our design does

not allow us to identify weather individuals preferences are stable. One reason is

that also the controls are affected by the monitoring by e.g. being colleague,

relative or a neighbor. More importantly, 25 per cent of our selected controls are

selected for monitoring in our follow up period.