29
6.4.2 Heterogeneity in effects depending on pre-retirement income and education
One other question is whether the reform effect differs across individuals by education
and pre-retirement income. The income loss from accepting the offer, in absolute
values, is lower for those with low pre-retirement incomes. If there is only a small
negative effect of reduced income on health, the positive effect on health could thus be
larger for those with low incomes. Another reason for a larger relative health effect
among the low-educated and those with low income, is that their work environments
and/or their health may be worse than those of high-educated/high-income individuals.
This potentially larger effect may be from the instant larger risk reduction, but it is more
likely to stem from overall poorer health of the low-educated at the time of being
offered retirement. That is, we have the context of a Grossman model in our minds
(Grossman 1971). Individuals with low education (low income) have had a more
physically demanding working career than those with high education (high income),
and consequently their health has deteriorated faster than that of the highly educated.
So, at age 55, low-educated individuals‟ health is worse than the health of those with a
higher level of education. However, this effect could also go in the other direction. The
reasoning for this is that individuals with a higher education and higher income may
value leisure more than individuals with less income and fewer outside options. That is,
increasing leisure time may increase life-satisfaction more for those with high
education, compared to those with less education, and thus better preserve the health
status of individuals with high education, and resulting in more positive health effects
for individuals with higher education.
In order to study whether there are any observed differences in effects, we
parameterized the intent-to-treat-model to be heterogeneous in both pre-retirement
income and education.
25
For the model that we allow to be heterogeneous with respect
to income, we assume the treatment effect to be linear in pre-retirement income; that is,
we specify the model as:
(
|
)
,
25
We have also estimated models where we have stratified on income and education respectively. The problem with
these estimations is the small sample sizes, which are why the precision is too low. The precision in the following
non-saturated models enables testing for differences in effects across income and education.
