32
retiring later. In the previous chapter we focused on the effects of the offer (i.e., the
intent-to-treat estimate). However, for comparison we also provided estimates with
regard to the effects of retirement on the number of days in inpatient care, using (1) a
cross-sectional OLS estimator and (2) a two-stage least squares (2SLS) estimator. For
the latter, the interaction term military and cohorts 1938-1939 is used as an instrument
for the actual retirement decision. Note that the 2SLS estimator allows us to estimate the
effect for the individuals who accept early retirement due to the Bill, what is known as
the compliers. Potentially this group of individuals differs in their response to early
retirement from the group of always-takers by having less severe health problems and
by having better work attachment. The implication is that the 2SLS estimator estimates
the local average treatment effect (the LATE) instead of the average treatment effect.
The results from these regressions are displayed in panels A and B in
The
results using OLS (panel A) show that early retirement is positively correlated with the
number of inpatient care days. These findings subsequently confirm results obtained in
other studies (see the introduction for references) using, for instance, cross section data
together with covariate adjustments in regression models. The results from 2SLS
28
clearly show that early retirement decreases the number of days in inpatient care over
the 15 years under study (from 56 to 70 years of age) by around 8.5 days. The major
part of the reduction occurs at later ages, i.e., between ages 61-70.
Table 10: Results from regressions (OLS and 2SLS) of number of days in inpatient care
on early retirement (prevalence of occupational pensions between the ages 55-59)
A: OLS
Ages 56-70
Ages 56-60
Ages 61-70
Effect
12.41**
(1.73)
14.27**
(1.74)
7.33**
(1.36)
7.98**
(1.37)
5.08**
(0.87)
6.29**
(0.88)
B: 2SLS
Effect
-8.48*
(3.93)
-13.34*
(6.50)
-2.57
(2.22)
-5.35
(4.58)
-5.92*
(2.80)
-7.99**
(3.02)
Controls
Yes
No
Yes
No
Yes
No
Notes: Excluded instrument in the 2SLS estimation is cohort 1938-1939*military. Robust standard errors in (): †
p<.1; * p<.05; ** p<.01. Each cell represents estimates from a separate model. All models include a military dummy
and dummy for cohort 1938-1939. Control variables (and interactions) are the same as in
Number of
observations is 19,986.
28
The results from the first step regression, already displayed in Figure 5, show that early retirement (occupational
pension in ages 55-59) increases by 67 percentage points for the group that was given the early retirement offer
compared to other groups (F = 949.11).
