30
where
is annual labor income at age 54, adjusted for prices, and
is 1 (0
else) if the individual has a college education. For the model that we allow for
heterogeneous effects from education, we instead specify:
(
|
)
.
The heterogeneous effect thus equals
( )
, where
x
is either
inc54
or
.
The results from the income and education heterogeneity analysis are shown in
columns (1) and (2) respectively of
These results indicate that the reform effect
is much stronger for low pre-retirement income employees, compared to employees
with high income before retirement (i.e., at age 54). In
the effect is evaluated at
three income levels: the 25
th
percentile, mean, and the 75
th
percentile (231,000 SEK
(26,837 Euro), 295,000 SEK (34,273 Euro), and 335,000 SEK (38,920 Euro),
respectively, all in the 2013 price level). The reform resulted in an effect on the number
of days in inpatient care, which is approximately 20 percentage points (or 50 percent)
larger for individuals with a pre-retirement income at the 25
th
percentile, compared to
individuals with a pre-retirement income at the 75
th
percentile.
As for education, the estimate of the reform effect is smaller for those with a college
education, in comparison to those without. For individuals who did not attend college,
the retirement offer reduced the number of days in inpatient care by more than 100
percent, while it reduced inpatient care days by about 34 percent for individuals with
high education.
26
To summarize, this analysis provides evidence of a larger effect of reduced number
of inpatient care days for those with low pre-retirement income or with low education.
The effects could stem from the larger negative income effect of those with higher
income (higher education), or from a more positive effect on health that stems from
poorer work environments and/or worse health among those with low income (lower
education).
26
In order to disentangle a potential separate effect from income and education, we specified a model in which both
education and income were integrated with the “treatment” and with each other. However, we could not find
significant effects with this specification. The problem is that there is not enough variation in the data to separately
identify the effects from income and education.
