Health Economics
- RESEARCH ARTICLE OPEN ACCESS
Incentives, Health, and Retirement: Evidence From a
Finnish Pension Reform
Joonas Ollonqvist1 | Kaisa Kotakorpi2,3 | Mikko Laaksonen4 | Pekka Martikainen5,6,7 | Jukka Pirttilä2,5,8 |
Lasse Tarkiainen5,7
1Finnish Institute for Health and Welfare, Helsinki, Finland | 2Finnish Centre of Excellence in Tax Systems Research (FIT), Tampere, Finland | 3Tampere
University, Tampere, Finland | 4Finnish Centre for Pensions, Helsinki, Finland | 5University of Helsinki, Helsinki, Finland | 6Max Planck Institute for
Demographic Research, Rostock, Germany | 7Max Planck – University of Helsinki Center for Social Inequalities in Population Health, Helsinki,
Finland | 8VATT Institute for Economic Research, Helsinki, Finland
Correspondence: Joonas Ollonqvist (joonas.ollonqvist@thl.fi)
Received: 28 March 2024 | Revised: 11 October 2024 | Accepted: 5 November 2024
Funding: This research was financially supported by NORFACE (Grant No. 462–14‐013), Yrjö Jahnsson foundation, the INVEST Research Flagship
(Research Council of Finland decision number: 345546), the Finnish Centre of Excellence in Tax Systems Research (FIT) (Research Council of Finland
decision numbers: 346250 and 346251), the European Research Council under the European Union's Horizon 2020 research and innovation programme
(grant agreement No 101019329), the Strategic Research Council (SRC) within the Research Council of Finland grants for ACElife (\#352543–352572) and
LIFECON (\#345219), and grants to the Max Planck—University of Helsinki Center from the Jane and Aatos Erkko Foundation, the Max Planck Society,
University of Helsinki, and Cities of Helsinki, Vantaa and Espoo.
Keywords: health | incentives | pension reform | retirement
ABSTRACT
This paper examines, using exogenous variation generated by a Finnish pension reform implemented in 2005, the interplay
between health and financial incentives to postpone retirement. Based on detailed administrative data on individual health and
retirement behavior, we focus on whether individual reactions to incentives vary according to health status and analyze whether
individuals with ill health are also able to take advantage of the potential monetary benefits of delayed retirement created by the
reform. We find that on average, individuals react to the financial incentives created by the reform as expected. This result holds
for most of the health‐related subgroups we analyze. However, those with a long period of sickness absence are less likely to
respond to changes in the financial incentives to postpone retirement.
JEL Classification: H55, J26
1 | Introduction
Many countries facedwith the challenge of population aging have
implemented pension reforms with the objective of extending
working lives and improving the sustainability of public finances
(OECD 2019). A key tool in these reforms has been to provide
financial incentives for individuals to postpone retirement.
In addition to the question of whether such incentives work on
average, an important and yet underexplored issue relates to
potential heterogeneity in individual responses, and the asso-
ciated equity implications of these types of reforms. This paper
focuses on one particular dimension of heterogeneity, which is
highly relevant for the elderly population, namely health.
Earlier empirical evidence indicates that ill health is a key factor
behind early old‐age retirement (e.g., van Rijn et al. 2014 and
Leijten et al. 2015). We examine whether incentives inherent in
pension reforms exacerbate this pattern: We ask whether such
incentives, and the associated monetary benefits, can only be
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly
cited.
© 2024 The Author(s). Health Economics published by John Wiley & Sons Ltd.
Health Economics, 2025; 34:537–572 537 of 598
https://doi.org/10.1002/hec.4917
utilized by persons who have relatively good health and are
therefore better positioned to continue working. Further,
because of a positive correlation between health and socioeco-
nomic status, the benefits of working longer may then be reaped
primarily by high‐income individuals. A new kind of efficiency‐
equity trade‐off may therefore arise: improving incentives could
increase the employment of older workers, but it may also in-
crease inequality among them if employment improvements
concentrate on healthy high socioeconomic status individuals.
While the impact of retirement incentives on the decision to
continueworking has been examined by a large number of earlier
studies—we discuss the literature below—ours is one of the first
papers on this potential trade‐off. We conduct our analysis in the
context of a Finnish pension reform implemented in 2005. This
was a major nationwide reform that influenced retirement in-
centives very differently across the population, depending for
example on the individual's age and accrued pension. The reform
therefore enables us to utilize exogenous variation in the in-
centives to retire. In addition to data on retirement behavior, we
utilize detailed administrative data on various individual health
measures. A key focus in this paper is the potential heterogeneity
in individual reactions to retirement incentives. If individuals
react differently to these incentives, and if the strength of the
reaction is correlated with one's health, the reform could indeed
lead to adverse equity consequences.
Overall, the way individuals react to incentives is a huge question
in economics. Individuals may differ in how theymake economic
decisions formany reasons, for example due to differences in time
preferences (Tanaka, Camerer, and Nguyen 2010), risk prefer-
ences (Gloede, Menkhoff, andWaibel 2015), liquidity constraints
(Carvalho,Meier, andWang2016) or self‐control (Bernheim,Ray,
and Yeltekin 2015). Potential differences in decision‐making in
terms of health (Decker and Schmitz 2016) and cognitive abilities
(Dohmen et al. 2010; Mani et al. 2013) have also been analyzed.
In our context, a connection between health and reactions to
incentives may arise for various reasons. First, individuals with
ill health may simply be unable to postpone retirement, and
their reactions to incentives may be muted for this reason.
However, many of the health conditions captured by our data
are such that they do not necessarily prohibit working per se.
Second, reacting to tax‐benefit policies in an optimal way often
requires difficult financial calculations and long‐term planning.
If an individual's attention is drawn to other problems associ-
ated with one's current life situation, the ability to engage in
long‐term planning may be hindered. Shah, Mullainathan, and
Shafir (2012) discuss this issue in the context of poverty; we
conjecture that a similar mechanism may be operational if an
individual is preoccupied with health problems. Finally, a factor
that points in the other direction is the following: to compensate
for the greater opportunity cost of working, individuals with ill
health may in fact require larger financial returns for continuing
to work and may therefore respond more strongly to improved
incentives. The pattern of heterogeneity that we should expect
to see is therefore not clear a priori.
Our results indicate that on average individuals react to retire-
ment incentives in the expected manner: The better the finan-
cial incentives to postpone retirement are, the more likely the
individual is to postpone retirement. Furthermore, many types
of individuals appear to react to retirement incentives, and it
therefore does not seem to be the case that the ability to take
advantage of better incentives is limited to any specific group.
On average, less healthy individuals retire earlier, as expected,
but our results do not indicate strong and consistent differences
in reactions to incentives between population groups defined
using health indicators related to having received treatment in
specialized care or purchases of prescribed medication. How-
ever, we do find that individuals with a spell of sickness absence
in the previous year react to incentives less strongly than other
groups. These results are important, given that sickness absence
is a health measure that is directly related to the ability to work.
The results suggest that individuals who have work related
health conditions for which they have not received effective
treatment—in the sense that the illness has lead to sickness
absenteeism—are less likely to respond to financial incentives to
postpone retirement than their more healthy peers.
In the main analysis, we focus on transitions to old‐age pension,
but the results are robust to including disability pension re-
cipients, an important group when considering the linkages
between health and retirement, into the analysis. Also a whole
host of other robustness checks are carried out and the main
results stay similar across specifications.
The analysis in this paper connects to the larger literature on the
effects of pension reforms on retirement behavior. The effects of
incentives on retirement have been studied for example in Fur-
geson, Strauss, and Vogt (2006), Coile and Gruber (2007),
Hanel (2010), Brown (2013), Johansson, Laun, and Palme (2014),
Manoli andWeber (2016), Hernæs et al. (2016) andEngels, Geyer,
and Haan (2017). Some of them find strong responses to retire-
ment incentives, but the results vary considerably. van Rijn
et al. (2014) and Leijten et al. (2015) studied the connection be-
tween health and retirement behavior. They found that ill health
increases the likelihood of early old‐age retirement, but these
papers did not analyze reactions to retirement incentives per se.1
Kerkhofs, Lindeboom, and Theeuwes (1999) estimate the impact
of incentives on the retirement decision, while at the same time
controlling for health, without examining the interaction be-
tween health and incentives. Staubli and Zweimüller (2013)
examine, in turn, the employment impacts of increases in the
early retirement age utilizing a policy reform inAustria. Theyfind
that the employment increases were greater among those with
better health (measured by sickness absence), whereas workers
with worse health found other pathways to retirement, such as
disability benefits. Their paper examines responses to age‐based
eligibility rules, rather than monetary incentives, which is in
our focus. A theory model of health and retirement is built by
Garcia‐Gomez et al. (2017). Their model predicts that wealthier
individuals (compared to poorer individuals) are more likely to
retire for health reasons and that health problems make older
workers more responsive to financial incentives. When exam-
ining empirically the interaction effect of health and retirement
incentives they focus on health shocks, defined as unpredicted
hospitalisations.
Our paper contributes to the literature in several ways.We exploit
exogenous changes created by a pension reform, and provide one
of the first studies offering a systematic analysis of potentially
538 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
heterogeneous reactions to the reform by people with different
health status. We use high‐quality administrative data on health,
covering a wide variety of health variables, such as mental health
and sickness absence. All our health indicators are objective
measures (as opposed to self‐reported health status) and lagged by
1 year, mitigating the potential endogeneity of health on labor
market outcomes. The benefit of working with a larger set of
health measures is that by doing so we can cover key dimensions
of health that are also significant determinants of retirement.
Earlier analysis of the Finnish reform of 2005 has been provided
by Uusitalo and Nivalainen (2013), who examined the mean
response to the reform and found relatively strong effects of the
incentives created by the reform on retirement decisions. They
did not analyze the potential heterogeneity in the reactions to
the reform, which is crucial for understanding the associated
efficiency‐equity trade‐offs. Leinonen et al. (2016) on the other
hand provide descriptive evidence showing that healthier in-
dividuals start to retire earlier after the reform. They, however,
do not attempt to provide causal estimates of the effects of the
reform or analyze how incentives affect retirement decisions at
the individual level. Gruber, Kanninen, and Ravaska (2022) also
analyzed the main effects of the reform, with a focus on the
effect of the change in the statutory retirement age, which was
another key feature of the reform. We control for the effect of
age limits in our analysis. The unique contribution of our study
is that we focus on differential reactions to incentives. In
particular, we provide a comprehensive assessment of the effects
of changes in retirement incentives on retirement decisions
brought about by a major national pension reform. We focus on
older workers, a population where health differences are
prominent, to assess heterogeneous effects of the reform on
different types of individuals.2
The rest of the paper is organized as follows. Section 2 introduces
the Finnish pension system and the 2005 reform. Section 3 pre-
sents the empirical strategy. Preliminary results are presented in
Section 4, and the robustness of the findings is discussed in Sec-
tion 5. Conclusions are offered in the final section.
2 | The Finnish Pension System and the 2005
Reform
The Finnish pension system has two elements: (1) earnings‐
related pensions and (2) residence‐based national pensions.
Participation in the earnings‐relatedpension system ismandatory
and covers virtually all earnings and workers. The level of an
individual's pension is determined by her working history, the
earnings received and age at retirement. National pensions and
the so‐called guarantee pension (introduced in 2011 to guarantee
aminimum level of income to all pensioners living in Finland) are
proportional to the earnings‐related pension. They are paid to
those individuals who have a low accrued pension. Each euro of
accrued earnings‐related pension cuts national pensions by 50
cents and the maximum amount of the national pension was
529.68 euro per month in 2005. Also, the marital status of the
individual affects the amount of the national pension.
The Finnish pension system has statutory retirement ages for
full and early old‐age pensions, explained in more detail below.
In addition to these two, there are several alternative retirement
paths, which differ in their eligibility criteria. The different
pathways include part‐time pension, disability pension, and
unemployment pension (abolished in the 2005 reform).
The reform implemented in 2005 implied major changes to the
key parameters of the pension system (age limits, accrual rates,
etc.). The first reform laws were passed in the middle of 2003,
and the new rules took effect as of January 2005. An informa-
tion campaign about the reform was implemented already at the
beginning of 2004.
The key feature of the reform for our analysis is that the new
rules implied different changes in the incentives to retire for
different groups of individuals. In short, the reform created
variation in retirement incentives between population groups
and over time, and this variation can be utilized to estimate the
causal effects of the reform. Below, we describe the changes
implemented in the reform in detail.
Table 1 shows the main features of the 2005 pension reform.
Before 2005, the full (or default) retirement age (FRA) was
65 years. Early old‐age retirement (ERA) was possible from age
60 onwards. Accrued pensions were cut by 0.4% for each month
of early retirement before the age of 65. Delaying retirement
after the age of 65 increased the accrued pension by 0.6% for
each month. The level of pensions was calculated based on
earnings for the last 10 years before retirement for most in-
dividuals, although the rules were different for job switchers.
The measure of accrued pension that we use in the analysis is
calculated by the Finnish Centre for Pensions, and accounts for
these complications. It was limited to 60% (66% for public‐sector
workers) of the highest annual salary for those years. Some
public sector workers also had different retirement ages
depending on their occupation. Pensions started to accrue at the
age of 23 and the accrual rate until age 59 was 1.5%. For 60‐ to
65‐year‐olds, the accrual rate was 2.5%. A so‐called halfway
index3 was used to convert accrued pensions to the retirement
year's money and to adjust pensions paid to those under
65 years old. For those over 65 years old, the earnings‐related
pension index was used to adjust the pension paid.4
The 2005 reform changed the fixed FRA to a flexible FRA. Since
2005 it has been possible to fully retire after the age of 63 and an
individual's entire working history is taken into account in
calculating the pension payments. The age limit of the ERA, on
the other hand, was increased to 62. Early retirement cut pen-
sions by 0.6% for each month of early retirement before the age
of 63. Postponing retirement after the age of 68 increased pen-
sions by 0.4% per month. In the reform, the eligibility age for the
full national pension remained the same, but the eligibility age
for the early national pension changed from 60 to 62. The
penalty for claiming the national pension early did not change.
In addition to age limits and the associated rules, accrual rates
changed as well. Following the reform, the accrual rate for the
18–52‐year‐olds was 1.5%. Between the ages 53 to 62, the accrual
rate was 1.9%. For 63‐ to 68‐year‐olds, the reform introduced a
so‐called “super” accrual rate of 4.5%, where the aim was to
encourage people to keep working after the minimum eligibility
retirement age. Further, pensions that accrued after the age of
539 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
63 no longer influenced an individual's national pension. The
halfway index was replaced by a wage coefficient5 and the re-
form abolished the differences in the regulations between public
and private sector workers.
3 | Data and Empirical Strategy
3.1 | Data and the Sample
We use individual‐level annual updated administrative data
from the Finnish Centre for Pensions and Statistics Finland. In
addition to crucial information on pensions and retirement
decisions, the data includes a large set of individual de-
mographic and labor market characteristics. In addition, the
data contain a wide range of individual health indicators based
on administrative data on hospital treatments, drug pre-
scriptions and sickness absence. The sources for the health data
are the Finnish Institute for Health and Welfare (THL) and the
Social Insurance Institution of Finland (Kela). The sample is an
11% random sample of all persons residing in Finland for at
least one year during 1987–2007 and the data for these in-
dividuals covers the years 2000–2009.
In the analysis, we concentrate on early and full old‐age retire-
ment, and we include individuals entitled to earnings‐related
pensions and/or national pensions. In Section 5, we check the
robustness of our findings by focusing on individuals receiving
only earnings‐relatedpensions, since they are themost affectedby
the changes in incentives caused by the reform.6
Our main sample includes private sector workers aged between
62 and 68 years (at the end of the year) who are in the labor
force. Being in the labor force is defined as not having retired
earlier. In one robustness check we run the analysis separately
for individuals who were or were not unemployed in the pre-
vious year. We focus on 62‐ to 68‐year‐old individuals, because
they are able to retire (early or full) both before and after the
reform. All workers other than private sector workers are
excluded from the sample, because of the inaccuracy of the data
and some differences in the accrual rules between sectors. Our
sample size is around 13,000 individual‐year observations.
We choose to exclude disability pensions from the main analysis
since we want to focus on retirement that is based more directly
on the decision of an individual. In Section 5, we examine the
sensitivity of our results to the way in which disability pensions
are handled, and show that our results are robust in this
respect too.
As mentioned above, the reform took effect in 2005, but full
information about the reform was available already in 2004. We
exclude the years 2004 and 2005 from our analysis to abstract
from potential anticipation effects.
3.2 | Measuring the Financial Incentives to Retire
We use the changes in total pension wealth, when retirement is
postponed by 1 year, to measure the financial incentives related
to retirement. A similar approach has been used for example in
Coile and Gruber (2007).7 We define pension wealth as the
present value of the stream of future pension incomes until age
one hundred. The benefit of using pension wealth instead of the
annual pension is that it measures the financial incentives of
postponing retirement more broadly. Analyzing changes in
pension wealth takes into account how postponing retirement
affects future pensions, as well as the fact that pension payments
are then received for 1 year less. Formally, pension wealth is
defined as:
PWr =∑
100
s=r
πsβs−rIs−rPr(r) (1)
In Equation (1) r indicates the age of retirement, Pr(r) is annual
pension including national pension (in year 2000 euros) when
TABLE 1 | 2005 pension reform.
Before reform After reform
Full retirement age (FRA) 65 63
Early retirement age (ERA) 60 62
Early claiming penalty for each month −0.4% −0.6%
Reference age for early claimin 65 63
Delayed claiming bonus for each month 0.6% 0.4%
Reference age for delayed claiming 65 68
Accrual %
Ages 18 to 22 1.5%
Ages 23 to 52 1.5% 1.5%
Ages 53 to 59 1.5% 1.9%
Ages 60 to 62 2.5% 1.9%
Ages 63 to 65 2.5% 4.5%
Ages 66 to 68 4.5%
Note: The reform maintained the eligibility age for the full national pension while adjusting the eligibility age for the early national pension from 60 to 62. There were no
changes made to the penalty for claiming the national pension early.
540 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
retired at age r, s is age, πs represents gender and age dependent
survival probability (from Statistics Finland) and β is the discount
factor (β = 0.97). After retirement, pensions are increased ac-
cording to an index I in real terms.8 In the calculations of I we use
the average annual growth rates of the consumer price index,
earnings index and national pension index from 1995 to 2015.9
Our measure for financial incentives is the expected relative
change in pension wealth when retirement is postponed by
1 year, based on the information available in period r:
Δ ln(PWr) = Er[ln(PWr+1)] − ln(PWr) (2)
We expect that when this measure increases, it becomes less
likely that the individual retires within a year. The notion in
Equation (2) is referred to as the “financial incentive to post-
pone retirement” in the remaining text, or sometimes simply
“incentive” for brevity. We calculate pension wealth for each
individual for every year using the accrual rules in place during
a particular year. In the calculations, we use information about
the individual's overall accrued pension at the end of 2004. The
source for this variable is the Finnish Centre for Pensions. We
further assume that the retirement date is always December 31.
The main reason is that individuals who are not retired at all do
not have any retirement date to use, and a fixed date within the
year ensures comparability. Furthermore, for the retirement
year and the year after retirement, we use earnings from the
year before retirement multiplied by the earnings index.
Figure 1 shows the distribution of the financial incentive to
postpone retirement for different individuals. The Figure dis-
plays the financial incentives in absolute euro terms for clarity
and transparency, while our econometric analysis below will
utilize relative incentive measures (as defined in Equation 2).
The figure indicates that for the majority of individuals, post-
poning retirement increases their pension wealth and on
average the increase is around 25,000 euros. However, the
financial incentive to postpone retirement varies substantially
among individuals. For some individuals, postponing retirement
decreases pension wealth, whereas for some their pension
wealth increases by over 80,000 euros.
To showcase how the incentives to postpone retirement were
altered in the reform, we next present results from a simulation
exercise. We first calculate pension wealth including both
earnings‐related pension and national pension10 and the in-
centives to postpone retirement for every individual with both
rules, and then compare these two. Since wealth and incentives
are calculated for each individual under both rules for the whole
study period, the only difference between the calculated values
is due to the changes in the rules. Note that in the actual
regression analysis, we do not only use a sample of the same
individuals in a panel, rather our data are in a repeated cross‐
section format.
Figure 2 illustrates how the reform changed the financial in-
centives to postpone retirement in euro terms. On average the
change in financial incentives to postpone retirement is slightly
negative, indicating that the reform on average mildly worsened
the financial incentives to delay retirement. However, the impli-
cations for incentives are not the same for every individual and
there is large variation as to how the reform changed incentives.
As was explained above, different age groups were affected
differently by the reform. Figure 3 shows how the reform
changed incentives by age groups (also in euros). According to
the Figure, the reform increased, on average, the incentives to
postpone retirement for 62‐ and 65‐year‐old individuals. For 63‐
and 64‐year‐olds, the incentives worsened on average. In
FIGURE 1 | The distribution of retirement incentives (EUR). The Figure displays the distribution of the change in the amount of pension wealth
when retirement is postponed by 1 year that is the financial incentive to postpone retirement, measured in euros. Pension wealth includes both
earnings‐related pensions and national pensions. The values are in year 2000 euros and extreme values are excluded. [Colour figure can be
viewed at wileyonlinelibrary.com]
541 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
addition, the reform implied a direct, overnight, increase in
pension wealth, driven by the reduction in the eligibility limit
for full old‐age pension from 65 to 63 years of age. These
changes are further described in Appendix B.
The reform created variation in incentives not only by cohort, but
also according to accrued pension and the level of earnings. This
variation in incentives is illustrated in Figure 4. On the x‐axis,
individuals are divided into percentiles according to the ratio
FIGURE 2 | The distribution of the change in retirement incentives (EUR) in the 2005 reform. The figure displays the distribution of the change
implied by the 2005 reform in the financial incentive to postpone retirement, measured in euros. Pension wealth includes both earnings‐related
pensions and national pensions. The values are in year 2000 euros and extreme values are excluded. [Colour figure can be viewed at
wileyonlinelibrary.com]
FIGURE 3 | The distribution of the change in retirement incentives (EUR) in the 2005 reform, by age group. The figure displays the distribution of
the change implied by the 2005 reform in the financial incentive to postpone retirement, measured in euros, by age group. Pension wealth includes
both earnings‐related pensions and national pensions. The values are in year 2000 euros and extreme values are excluded. [Colour figure can be
viewed at wileyonlinelibrary.com]
542 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
between earnings in the following year and current‐year total
pension. The y‐axis shows the change in the financial incentive to
delay retirement, implied by the reform. As can be seen from
Figure 4, the incentives to postpone retirement improved for
those with high earnings compared to their accrued pension and
worsened for those with low earnings compared to their accrued
pension. The change in financial incentives implied by the reform
also clearly increases and is roughly monotonous over the per-
centiles, but the increase is not linear.
3.3 | Measurement of Health
We measure health with several different variables, derived
from various national registers containing detailed individual‐
level information on different aspects of health.11
First, we use hospitalization and treatment information from
the hospital discharge records of the Finnish Institute for
Health and Welfare (THL). Using this register, we form an
indicator of whether or not an individual has been treated in
hospital for any reason. We also analyze cardiovascular dis-
eases and diseases of the musculoskeletal system and connec-
tive tissue separately, with the identification of these conditions
based on the International Classification of Diseases and
Causes of Death (ICD‐10).12
Second, we use medication information from the reimburse-
ment register of the Social Insurance Institution of Finland. We
form two variables using data on purchases of prescription
medications. The first measure is formed according to total
purchases of any medication within a year and is divided into
three categories (fewer than 4 purchases, 4 to 7, and 8 or more).
The rationale for using the number of purchases as a proxy for
health problems is the following: according to Finnish health
insurance law, one can purchase 3 months worth of subscrip-
tion medication, subject to reimbursement from public health
insurance, at any one time. Therefore, four purchases within a
year is a proxy for a treatment need of a chronic condition
associated with particular medication. Further purchases signify
that the individual is likely treated for multiple conditions or
takes several different types of medication. The second measure
relates specifically to mental health, where we use information
on purchases of prescribed psychotropic medication to form an
indicator of mental health problems.13
Third, we use information on sickness absence from the Social
Insurance Institution of Finland. We measure sickness absence
as the total length of all sickness absence spells in days within a
year. Sickness allowances, however, are paid only after a spec-
ified (usually 10 days) waiting period and thus the register only
contains information about those sickness absences that are
longer than the waiting period. We divide the length of sickness
absences (days after the waiting period) into four categories
(0 days, 1–14 days, 15–60 days, and over 60 days).14
In addition to these measures, we form a more general indicator
of ill health, dividing the sample into sub‐samples according to
overall health status. We describe the formation and rationale
behind this measure when we carry out the sub‐sample analysis.
A couple of notes on the nature of our health data are in order.
First, regarding sectoral coverage, the hospital discharge records
relate to use of specialized health care, and these data have very
good coverage of both publicly and privately provided overnight
hospital care services. Coverage of outpatient care in specialized
healthcare is limited, on the other hand. Therefore these data
likely provide information on relatively severe cases of ill health.
FIGURE 4 | The change in retirement incentives in the 2005 reform, by the ratio of future earnings and pension. The figure displays the change in
the financial incentive to postpone retirement implied by the 2005 reform, by the ratio of future earnings and pension. Pension wealth includes
earnings‐related pensions and national pensions. [Colour figure can be viewed at wileyonlinelibrary.com]
543 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
More information on the coverage of these data is provided in
Sund (2012).15 However, our other health measures cover all
sectors of health care provision and information on medication
purchases is thus available regardless of where the medication
was prescribed (primary or secondary care; and public, private,
or occupational healthcare). In addition, information on sick
leave does not hinge on where the leave was prescribed.
Second, regarding measurement of health using administrative
data, the hospital discharge data and the prescription data cover
only health problems for which individuals have sought treat-
ment. This is a common feature when using administrative data
to measure individual health. All purchases of prescription
medication in Finland are from authorized pharmacies by pre-
scription from a medical doctor after clinical assessment. Thus,
our health measures, particularly those based on hospital use,
are likely to capture more severe health problems.
Third, sickness absence is a particularly important health mea-
sure for our purposes, for two reasons: This variable covers cases
where prior treatments—whether sought on time or not—have
not been effective in maintaining the individual's working capa-
bilities. Further, this measure directly captures the labor market‐
relevant aspects of health, and is of particular interest when
analyzing individual choices between work versus retirement.
Indeed, as we show below, this is the health measure in our data
that is most strongly correlated with the likelihood of retirement
in a given year.
3.4 | Descriptive Statistics
Table 2 shows some descriptive statistics for our main sample.
The information is provided separately for the periods before
and after the reform. While there are variables such as age and
gender that are not expected to change over the period, many
others are either directly or indirectly influenced by the reform,
such as pension (wealth) levels. Also, general time trends may
influence the health‐related variables.
Approximately 20%–30% of the target group retires each year,
and a much larger share (32% instead of 14%) reaches the full
retirement age after the reform, due to a lowering of the full
retirement age. Individuals have approx. 35 years of work his-
tory on average, and the mean pension is close to 15,000 euros a
year in comparison to mean annual earnings of around 24,600
euros. When it comes to the measures of health, a small share
receive treatment for a specific type of illness. Many individuals
(around 12%) have sickness absences and use a large amount of
prescription medication. For example, 7%–9% of the individuals
in the sample used psychotropic medication.
TABLE 2 | Means of selected variables before and after the reform.
Before the reform After the reform
Retirement rate 0.202 0.295
Reaching full retirement age 0.144 0.321
Female 0.430 0.395
Age (at the end of the year) 63.13 62.96
Spouse, share 0.706 0.715
Working history (years) 34.87 37.08
Unemployed, share (t − 1) 0.0504 0.0966
Tertiary degree, share 0.0998 0.137
Earnings 21,518 26,037
Accrued pension (euros) 13,132 16,851
Pension wealth (euros) 279,220 348,091
Pension wealth (logs) 12.43 12.63
Δ ln(PW) 0.0659 0.0678
Δ (PW) 20,379 24,855
Psychotropic medication (t − 1) 0.0681 0.0917
Medication purchases (t − 1) 5.688 7.022
Any treatment (t − 1) 0.106 0.114
Treatment, cardio (t − 1) 0.0228 0.0217
Treatment, muscular (t − 1) 0.0249 0.0236
Sickness absences days (t − 1) 5.170 4.909
Sickness absences share (t − 1) 0.127 0.115
Note: Monetary values are in year 2000 euros. Δ ln(PW) is the financial incentives to postpone retirement (i.e. the relative change in pension wealth, PW , when
retirement is postponed by 1 year). Sickness absences in days refers to the length of sickness absence spells after the waiting period. The share of unemployed in
the previous year is identified by using information about the main type of activity of an individual within a year.
544 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
3.5 | Estimation Strategy
Our main question is how financial incentives affect retirement
behavior and whether the reactions vary between individuals
with different health status. Our analysis includes the years
around the 2005 reform: years 2001–2003 (before) and 2006–2008
(after the reform). The years 2004 and 2005 are dropped to abstract
from possible anticipation effects related to the reform.
We exploit variation in the financial incentives to postpone
retirement between population groups, created by the reform, to
study how incentives affect retirement. To be precise, we esti-
mate the following regression:
Ri,t = θ1Δ lnPWi,t) + θ2 lnPWi,t) + γt + βXi,t + ui,t (3)
where Ri,t is an indicator variable equal to one if individual i
retires in year t, conditional on not having retired earlier. The
time fixed effect is denoted by γ. The vector X contains indi-
vidual‐level control variables, such as gender, age, work history,
and education, spouse controls and health indicators.16 PWt is
an individual's calculated pension wealth at the end of year t
and Δ lnPWi,t) captures the incentives to postpone retirement.
The parameter of interest is θ1, which captures the effect of
financial incentives on the retirement probability, while the
control variables (including the actual level of pension wealth)
capture some of the other determinants of retirement behavior.
To highlight the key features of our empirical strategy, our main
specification resembles a difference‐in‐difference type analysis:
We utilize data from years before and after the reform, and the
reform caused financial incentives to change differently for
different population groups. These changes were exogenous that
is it is not possible for individuals to select into a particular
group; we however note one potential caveat to this feature
below. The regression includes controls for the relevant char-
acteristics of those groups (similar to group dummies in a DiD
analysis). Instead of having one treatment dummy, in our
setting treatment intensity varies according to how much in-
centives changed in the reform.
After analyzing average responses, we study how the reactions
to incentives vary by health status. We choose to present models
estimated using different sub‐samples, because we favor an
approach that allows the importance of other determinants of
retirement decisions to vary across different population groups.
This approach is more flexible and easier to interpret than the
alternative of interacting the group indicators with the incentive
measures. Nevertheless, we assess the robustness of our findings
by also estimating a unified model that incorporates interaction
terms between the incentives, pension wealth, reaching full
retirement age, and the health variables.
As explained above, our main empirical strategy utilizes the
pension reform as a source of exogenous variation in retirement
incentives. Most individual characteristics that influence the
financial incentives to postpone retirement and the way the
reform changed those incentives (such as accrued pension,
gender, and age) are exogenous near retirement. However,
accrual rates and hence our measure of financial incentives
depend on earnings, as shown in Figure 4. Earnings levels on
the other hand may reflect changes in labor supply, or under-
lying differences in the preference for working, and this part of
the variation in the incentive measure may be partially
endogenous.
Therefore, as a robustness check, we complement our main
analysis by implementing an IV analysis. Here, we follow an
approach often used in estimating individual responses to tax
changes, introduced by Gruber and Saez (2002). In order to
isolate only the exogenous variation in incentives caused by the
reform, the idea is to construct the incentive measure for each
year using the accrual rules in place in that year, but calculating
them with lagged earnings instead of current earnings.
We present the IV results only after the baseline OLS analyses as
a robustness check. Reassuringly, the results are very similar,
and the potential endogeneity therefore appears to play little
role in our case.
It will be helpful to introduce some additional notation to
explain the instrument used in more detail. The key regressor is
the financial incentive to postpone retirement,
Δ ln(PWt) = ln(PWt+1) − ln(PWt). Let us denote the log of the
pension wealth by lnPWt yt,…)), because the pension wealth
depends, among other things, on the earnings level, y, and the
pension wealth is calculated using the policy rules of year t.
Now the instrument for lnPWt+1 yt+1,…)) − lnPWt yt,…)) is
lnPWt+1 yt,…)) − lnPWt yt−1,…)).17 The exclusion restriction is
that individuals cannot affect previous earnings and therefore the
only variation in the incentives come from the policy change,
which is exogenous from the viewpoint of the individuals.
In addition to these main analyses, we carry out a wide variety
of additional robustness checks, as outlined in Section 5.
4 | Results
4.1 | Results for the Full Sample
We first provide some graphical evidence to illustrate how
changes in the financial incentives to postpone retirement, due
to the reform, are related to the corresponding change in the
probability of retirement. This is depicted in Figure 5. The x‐axis
shows how the reform changed the incentives in relative terms
among various groups, whereas the y‐axis shows how the
probability of retirement changed in the same groups around
the reform. The groups are formed according to age, gender and
work history (less than 35 years, 35–40 years, at least 40 years)
and the numbers in the graph show the age of each group.
Clearly there is a negative connection between the incentives to
postpone retirement and the retirement probability.
Next, we report the mean impact of the reform on retirement
using the specification in Equation (3). These results are shown in
Table 3, wherewe successively addmore control variables in each
545 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
column. The first column reports results from the regressionwith
year fixed effects, while the second column adds the level of
pension wealth to the model. The third model contains basic
individual‐specific control variables,which include also a dummy
measuring whether the person has reached the full retirement
age.18 Controlling for this factor is important, given that reaching
the default retirement age has been found to be a crucial deter-
minant of retirement decisions (Gruber, Kanninen, and Rav-
aska 2022). The fourth model also includes controls for an
individual's spouse. In models (5) to (10), the health variables are
included separately one at a time, and in model (11) all the indi-
vidual health variables are included simultaneously.
The signs of the estimates are as expected. The coefficient of the
log change in pension wealth, Δ ln(PW), is negative, indicating
that when financial incentives to postpone retirement are
increased, individuals are indeed less likely to retire during the
analysis year. On the other hand, the level of accrued pension
wealth itself (ln(PW)) is associated with a higher likelihood of
retirement. These results are stable across the different specifi-
cations. With the full set of controls (Column 11), the estimated
coefficient indicates that a 1% point increase in the financial
incentives decreases the risk of retirement by around 2.2%
points. A one‐unit increase in the level of log of pension wealth,
in turn, increases the risk of retirement by 4.8% points (Column
11). Compared to earlier findings from Finland (Uusitalo and
Nivalainen 2013), our estimation results for the effect of in-
centives have the same sign, but the magnitude is smaller,
which may be due for example to the fact that Uusitalo and
Nivalainen (2013) conduct the analysis at a group level. In
Sweden, Johansson, Laun, and Palme (2014) find rather similar
results on the effects of incentives as we do. In sum, individuals
respond to retirement incentives as expected, and the positive
association between the level of wealth and retirement is
consistent with the notion that leisure time is a normal good
(higher wealth levels lead to earlier retirement).
As in Gruber, Kanninen, and Ravaska (2022), we find that
reaching the full retirement age (FRA) is an important deter-
minant of the retirement decision. The estimated coefficient on
the dummy indicating that an individual has reached the stat-
utory retirement age is around 0.24 across all specifications.
Regarding the health variables, in many cases worse health is
associated with a higher risk of retirement. Psychotropic
medication increases the risk of retirement by 4.1% points and
having at least 8 prescription medication purchases increases it
by around 2.2% points. Sickness absence increases the risk of
retirement as well. Having a spell of 1–14 sickness absence days
(after the waiting period) increases the risk by around 3.3%
points and over 60 sickness days increases it by around 7.3%
points.19 When the health variables are included simulta-
neously, the use of psychotropic medication and having had 1 to
14 or at least 60 sickness absence days remain statistically sig-
nificant determinants of retirement behavior. In particular, long
spells of sickness absence are, quite intuitively, important de-
terminants of retirement behavior.
4.2 | Results by Subgroups
Next, we analyze how the results vary between different types of
individuals with varying health. The heterogeneity analyses are
conducted by running regressions using separate sub‐samples.
The samples are divided according to whether an individual has
(i) had any treatment in a hospital; (ii) received treatment for
cardiovascular diseases or (iii) musculo‐skeletal diseases; (iv)
FIGURE 5 | Relationship between the change in retirement probability and the change in retirement incentives in the 2005 reform. The figure
displays the relationship between the change in the financial incentive to postpone retirement (x‐axis) implied by the 2005 reform, and the
change in retirement probability (y‐axis), for different groups. The groups are formed according to age, gender, and length of work history (three
categories). The displayed age represents the age of the group at the end of the year. The red line is the regression line, which is weighted by
group size. [Colour figure can be viewed at wileyonlinelibrary.com]
546 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TA
BL
E
3
|
Ba
se
lin
e
O
LS
es
tim
at
es
.
(1
)
(2
)
(3
)
(4
)
(5
)
(6
)
(7
)
(8
)
(9
)
(1
0)
(1
1)
D
ep
en
de
nt
va
ria
bl
e
is
th
e
re
tir
em
en
td
ec
isi
on
an
d
m
ea
n
re
tir
em
en
tr
at
e
in
ea
ch
m
od
el
is
0.
26
5
Δ
ln
(P
W
)
−3
.9
29
**
*
−4
.1
57
**
*
−2
.2
16
**
*
−2
.2
27
**
*
−2
.2
26
**
*
−2
.2
29
**
*
−2
.2
27
**
*
−2
.2
25
**
*
−2
.2
25
**
*
−2
.2
43
**
*
−2
.2
44
**
*
(0
.1
26
)
(0
.1
29
)
(0
.1
26
)
(0
.1
26
)
(0
.1
26
)
(0
.1
26
)
(0
.1
26
)
(0
.1
26
)
(0
.1
26
)
(0
.1
26
)
(0
.1
26
)
ln
(P
W
)
0.
10
3*
**
0.
04
89
**
*
0.
04
82
**
*
0.
04
81
**
*
0.
04
82
**
*
0.
04
82
**
*
0.
04
83
**
*
0.
04
79
**
*
0.
04
81
**
*
0.
04
80
**
*
(0
.0
08
16
)
(0
.0
10
8)
(0
.0
10
8)
(0
.0
10
8)
(0
.0
10
8)
(0
.0
10
8)
(0
.0
10
8)
(0
.0
10
8)
(0
.0
10
8)
(0
.0
10
8)
FR
A
0.
23
8*
**
0.
23
7*
**
0.
23
7*
**
0.
23
7*
**
0.
23
7*
**
0.
23
7*
**
0.
23
7*
**
0.
23
7*
**
0.
23
8*
**
(0
.0
12
7)
(0
.0
12
7)
(0
.0
12
7)
(0
.0
12
7)
(0
.0
12
7)
(0
.0
12
7)
(0
.0
12
7)
(0
.0
12
7)
(0
.0
12
7)
Tr
ea
tm
en
t,
ca
rd
io
(t
−
1)
0.
01
70
0.
01
09
(0
.0
22
8)
(0
.0
25
6)
Tr
ea
tm
en
t,
m
us
cu
la
r(
t−
1)
0.
03
56
0.
02
46
(0
.0
22
3)
(0
.0
25
4)
A
ny
tre
at
m
en
t(
t−
1)
0.
00
79
4
−0
.0
22
3
(0
.0
10
5)
(0
.0
13
9)
Ps
yc
ho
tro
pi
c
m
ed
ic
at
io
n
(t
−
1)
0.
04
13
**
*
0.
03
24
*
(0
.0
12
5)
(0
.0
12
8)
M
ed
ic
at
io
n
pu
rc
ha
se
s
(t−
1)
4
to
7
−0
.0
02
70
−0
.0
05
65
(0
.0
08
52
)
(0
.0
08
55
)
8
or
m
or
e
0.
02
19
**
0.
01
39
(0
.0
07
55
)
(0
.0
07
86
)
Si
ck
ne
ss
ab
se
nc
es
(t−
1)
1–
14
da
ys
0.
03
30
*
0.
03
35
*
(0
.0
16
0)
(0
.0
16
5)
15
–6
0
da
ys
0.
01
95
0.
02
17
(0
.0
14
8)
(0
.0
15
9)
O
ve
r6
0
da
ys
0.
07
30
**
*
0.
06
97
**
(0
.0
22
1)
(0
.0
23
7)
O
bs
er
va
tio
ns
12
,9
77
12
,9
77
12
,9
23
12
,9
23
12
,9
23
12
,9
23
12
,9
23
12
,9
23
12
,9
23
12
,9
23
12
,9
23
R‐
sq
ua
re
d
0.
08
7
0.
09
9
0.
31
8
0.
31
9
0.
31
9
0.
31
9
0.
31
9
0.
32
0
0.
32
0
0.
32
0
0.
32
1
FE
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
(C
on
tin
ue
s)
547 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
had 8 or more purchases of medication; (v) had medication for
mental illnesses; and (vi) had at least one sickness absence day
after the waiting period.
Finally, we form an indicator of ill health which combines in-
formation from the different health measures. This combined
measure is a complement to the individual measures, and at-
tempts to capture the individuals' health status more broadly.
The indicator gets value one if the individual has been an
inpatient due to cardiovascular diseases in a particular year, has
8 or more purchases of prescription medication, has psycho-
tropic medication or has over 60 days of sickness absence. The
rationale behind this indicator is to make the two groups—more
and less healthy persons—more equal in terms of their size.
Our interest here is in the equity effects of the reform, in
particular the interaction between health inequality and eco-
nomic inequality. We would like to examine whether the reform
has a differential effect on those who are prone to retire earlier
due to health problems in the baseline. Do those individuals who
have a larger risk of retiring early react to incentives ‐ and
correspondingly, are they able to utilize the potential financial
benefits created by the reform? It is possible, of course, that the
individuals differ in other respects than health as well, but we
attempt to mitigate this issue by always including a large number
of control variables (also related to the spouse).
The purpose of the subgroup analysis is therefore to detect
whether there is a risk that providing incentives for continuing
to work have ramifications in terms of aggravated inequality
among older workers. However, it should also be noted that it is
not necessarily clear that those with worse health are less in-
clined to react to incentives: it is also possible that they in fact
require stronger financial returns for continuing to work to
compensate for the greater opportunity cost of working. The
pattern of heterogeneity that we should expect to see is therefore
not clear a priori, making the empirical subgroup analysis all
the more important and interesting. Further, the link between
health and the reaction to incentives may of course differ
depending on the type of the underlying health issues.
The results for the subgroup analysis, using our main specifi-
cation corresponding to Equation (3) with the full set of con-
trols, are reported in Table 4. Overall, those with different types
of health problems mostly do react to the incentives and to the
level of pension wealth in a similar way as individuals in our
sample overall. This is revealed by the test statistics indicating
whether the estimated coefficient of the financial incentives
differ in a statistically significant way between the two groups
studied. This holds for most health indicators. An important
finding, therefore, is that many types of individuals do appear to
react to retirement incentives.
However, the reactions of individuals with sickness absence
spells (exceeding the 10‐day deductible period) differ from those
of other individuals in our sample: individuals with an extended
period of sickness absence react less to the financial incentives
to postpone retirement than others do. Furthermore, they react
less to reaching the full retirement age. The finding that we see
differential reactions for this group in particular is of interest,
given that sickness absence is a health indicator with theTA
BL
E
3
|
(C
on
tin
ue
d)
(1
)
(2
)
(3
)
(4
)
(5
)
(6
)
(7
)
(8
)
(9
)
(1
0)
(1
1)
In
di
vi
du
al
co
nt
ro
ls
N
O
N
O
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
Sp
ou
se
co
nt
ro
ls
N
O
N
O
N
O
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
H
ea
lth
co
nt
ro
ls
t−
1
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
YE
S
N
ot
e:
Δ
ln
(P
W
)
is
th
e
fin
an
ci
al
in
ce
nt
iv
es
to
po
st
po
ne
re
tir
em
en
t(
i.e
.t
he
re
la
tiv
e
ch
an
ge
in
pe
ns
io
n
w
ea
lth
,P
W
,w
he
n
re
tir
em
en
ti
s
po
stp
on
ed
by
1
ye
ar
).
Si
ck
ne
ss
ab
se
nc
es
in
di
ca
te
s
da
ys
af
te
r
th
e
w
ai
tin
g
pe
rio
d.
Ro
bu
st
sta
nd
ar
d
er
ro
rs
in
pa
re
nt
he
se
s
**
*p
<
0.
00
1,
**
p
<
0.
01
,*
p
<
0.
05
.
548 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TA
BL
E
4
|
O
LS
es
tim
at
es
by
po
pu
la
tio
n
gr
ou
ps
.
Tr
ea
tm
en
t,
ca
rd
io
(t
−
1)
Tr
ea
tm
en
t,
m
us
cu
la
r
(t
−
1)
A
ny
tr
ea
tm
en
t(
t−
1)
M
ed
ic
at
io
n
(t
−
1)
Ps
yc
ho
tr
op
ic
m
ed
ic
at
io
n
(t
−
1)
Si
ck
ne
ss
ab
se
nc
es
(t
−
1)
Ill
he
al
th
(t
−
1)
Ye
s
N
o
Ye
s
N
o
Ye
s
N
o
8≤
7
or
le
ss
Ye
s
N
o
1≤
N
o
Ye
s
N
o
D
ep
en
de
nt
va
ria
bl
e
is
re
tir
em
en
td
ec
isi
on
Δ
ln
(P
W
)
−3
.1
08
**
−2
.2
07
**
*
−2
.8
43
**
*
−2
.2
20
**
*
−2
.3
63
**
*
−2
.2
25
**
*
−2
.1
53
**
*
−2
.2
67
**
*
−2
.3
42
**
*
−2
.2
18
**
*
−1
.1
05
**
−2
.3
52
**
*
−2
.2
93
**
*
−2
.1
90
**
*
(1
.0
01
)
(0
.1
27
)
(0
.8
50
)
(0
.1
28
)
(0
.3
65
)
(0
.1
34
)
(0
.2
08
)
(0
.1
59
)
(0
.3
99
)
(0
.1
33
)
(0
.4
18
)
(0
.1
32
)
(0
.1
95
)
(0
.1
66
)
ln
(P
W
)
−0
.1
26
0.
05
06
**
*
0.
05
11
0.
04
85
**
*
0.
02
54
0.
05
16
**
*
0.
05
28
**
0.
04
26
**
0.
06
13
0.
04
71
**
*
0.
04
22
0.
04
93
**
*
0.
05
16
**
0.
04
32
**
(0
.0
80
5)
(0
.0
11
0)
(0
.0
93
7)
(0
.0
10
9)
(0
.0
34
2)
(0
.0
11
4)
(0
.0
18
3)
(0
.0
13
5)
(0
.0
37
8)
(0
.0
11
3)
(0
.0
34
9)
(0
.0
11
4)
(0
.0
17
5)
(0
.0
13
9)
FR
A
0.
25
3*
0.
23
7*
**
0.
04
53
0.
24
0*
**
0.
18
5*
**
0.
24
2*
**
0.
25
4*
**
0.
22
7*
**
0.
21
9*
**
0.
23
7*
**
0.
15
6*
**
0.
24
8*
**
0.
25
3*
**
0.
22
7*
**
(0
.1
01
)
(0
.0
12
9)
(0
.1
14
)
(0
.0
12
8)
(0
.0
42
3)
(0
.0
13
4)
(0
.0
21
9)
(0
.0
15
7)
(0
.0
47
6)
(0
.0
13
3)
(0
.0
37
8)
(0
.0
13
5)
(0
.0
20
6)
(0
.0
16
2)
p‐
va
lu
e
Δ
ln
(P
W
)
0.
32
1
0.
42
7
0.
71
8
0.
66
2
0.
76
1
0.
00
4
0.
68
5
p‐
va
lu
e
ln
(P
W
)
0.
01
6
0.
97
5
0.
45
9
0.
65
2
0.
71
2
0.
84
5
0.
70
7
p‐
va
lu
e
FR
A
0.
86
1
0.
06
0.
19
1
0.
30
3
0.
70
5
0.
02
0.
30
7
O
bs
er
va
tio
ns
28
7
12
,6
36
31
3
12
,6
10
14
47
11
,4
76
45
01
84
22
10
88
11
,8
35
15
43
11
,3
80
49
90
79
33
R‐
sq
ua
re
d
0.
43
8
0.
31
9
0.
41
2
0.
32
0
0.
32
0
0.
32
2
0.
31
2
0.
32
7
0.
31
2
0.
32
4
0.
34
3
0.
32
0
0.
31
3
0.
32
7
FE
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
In
di
vi
du
al
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
Sp
ou
se
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
H
ea
lth
co
nt
ro
ls
t‐1
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
ot
e:
Th
en
ul
lh
yp
ot
he
sis
fo
rt
he
p‐
va
lu
es
is
th
at
th
ed
iff
er
en
ce
be
tw
ee
n
th
ec
oe
ffi
ci
en
ts
eq
ua
ls
ze
ro
.Δ
ln
(P
W
)
is
th
efi
na
nc
ia
li
nc
en
tiv
es
to
po
stp
on
er
et
ire
m
en
t(
i.e
.t
he
re
la
tiv
ec
ha
ng
ei
n
pe
ns
io
n
w
ea
lth
,P
W
,w
he
n
re
tir
em
en
t
is
po
stp
on
ed
by
1
ye
ar
).
Si
ck
ne
ss
ab
se
nc
es
in
di
ca
te
s
da
ys
af
te
r
th
e
w
ai
tin
g
pe
rio
d.
Ro
bu
st
sta
nd
ar
d
er
ro
rs
in
pa
re
nt
he
se
s
**
*p
<
0.
00
1,
**
p
<
0.
01
,*
p
<
0.
05
.
549 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
clearest a priori link to labor‐market behavior. Having an
extended period of sickness absence also points to untreated or
unmanaged health problems, as we discussed in Section 3.3.
The information provided in Table A3 sheds light on the issue
how individual characteristics differ between those with or
without sickness absence. People with sickness absence have,
indeed, much more health issues than the persons without
sickness absence also according to the other health indicators.
For instance, those with sickness absence have had 10 times
more often at least some treatment in the previous year. The
sickness absence variable may thus best capture the least
healthy group of people. These individuals are also less well off
than others for example they are less well educated and have a
lower level of accrued pension wealth.
To summarize, the results indicate that economic incentives and
the level of pension wealth as well as health status matter for
retirement decisions, and many types of individuals, with
different health status, do appear to react to incentives. How-
ever, since those with sickness absence react less strongly to
economic incentives, this group may need special attention
when planning pension policies. For instance, some individuals
in this group might benefit from similar treatment as those
actually covered by disability insurance; in other words, access
to retirement before the rest of the population.
An interesting question relates to the comparison of our results
with those of Leinonen et al. (2016), who found that relatively
healthier individuals started to retire earlier after the reform.
This finding is not inconsistent with our results. First, one
should check whether the reform changed incentives differently
according to health. However, the figures in Appendix: The
Effect of the Reform on Incentives According to Health Statuses
indicate no systematic differences in this respect. One potential
explanation for the finding in Leinonen et al. (2016), which is
consistent with what we find, is that according to our results
healthy individuals appear to react somewhat more strongly to
the statutory full retirement age, which also changed at the
reform (c.f., Table 4). While it is difficult to pinpoint the precise
reason behind this difference, one potential explanation is that
less healthy individuals are more likely to have already retired
before reaching the full retirement age.
5 | Robustness
5.1 | Earnings‐Related Pensions
In the analysis above we included individuals receiving earnings‐
related pension and/or national pension. However, the reform
mainly concerned the earnings‐related pension system, and in-
dividuals receiving only earnings‐related pension were affected
more than individuals receiving national pension. In addition,
individuals receiving only earnings‐related pension have overall
higher levels of pension wealth. These two observations together
indicate that there might be differences in responses between
these two types of individuals.
For this reason, we also examined the recipients of earnings‐
related pension only, excluding those individuals who received
any amount of national pension. The results of the full model
(corresponding to the specification reported in the last column
of Table 3 for the full sample) are reported in the first column of
Table 5. The impact of financial incentives to postpone retire-
ment remains significant for this smaller sample as well, but the
magnitude of the coefficient rises. This is to be expected, since
the changes in the incentives are muted for those receiving
national pension.
We also conduct the robustness analysis by partitioning the data
into subgroups (see Table A6). The results tell a similar story as
the main analysis. Most subgroups' retirement decisions react to
the financial incentives to postpone retirement. Furthermore,
individuals with sickness absence react less to incentives than
other individuals.
5.2 | Unemployed Individuals
Unemployed individualsmayhave different reasons to retire than
employed individuals simply because they do not have a job to
continue in. Furthermore, theremight be differences in the usage
of sickness absence between employed and unemployed in-
dividuals.20 Therefore we also run the analysis excluding in-
dividuals who were unemployed in the previous year from the
sample.We have information about themain type of activity of an
individual within a year and use that to identify unemployed
individuals.
The baseline estimation results are shown in the second column
of Table 5. The estimated coefficient for the effect of the in-
centives is much smaller than with the main specification, but
still highly significant. As excepted, having at least one sickness
benefit spell has a larger impact on the risk of retirement in this
sub‐sample. On the other hand, if one only includes the un-
employed (the third column in Table 5), the point estimate of
the incentives strongly increase. The unemployed are therefore
a group for whom the financial incentives to postpone retire-
ment matter a great deal. A potential explanation is that a key
difference between being unemployed versus retired is indeed
the type of monetary compensation or social benefit received,
rather than the amount of labor supply. Therefore other reasons
that affect the labor supply versus retirement decision are likely
less relevant for this subgroup, leading to a more pronounced
role of monetary incentives.
The population subgroup results for the subgroup that excludes
the unemployed are displayed in Table A7. The results are fairly
similar to our main analysis, in that different types of individuals
react to incentives to postpone retirement. In this subgroup, dif-
ferences in reactions to incentives between individuals with and
without a previous spell of sickness absence are qualitatively
similar to the main analysis, but the difference is not statistically
significant.
5.3 | Disability Pensions
In the main analysis, we excluded disability pensions because
we wanted to concentrate on retirement that is more directly
based on the active decision of an individual. This choice may
550 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TABLE 5 | Robustness of the baseline results.
OLS 2SLS
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent variable is retirement decision
Δ ln(PW) −2.680*** −1.335*** −3.730*** −2.174*** −2.305*** −2.380*** −2.864***
(0.176) (0.137) (0.598) (0.128) (0.125) (0.129) (0.146)
ln(PW) 0.0628*** 0.0422*** 0.184*** 0.0466*** 0.0500*** 0.049*** 0.046***
(0.0158) (0.0112) (0.0407) (0.0110) (0.0107) (0.011) (0.011)
Δ ln(pension) −2.243***
(0.126)
ln(pension) 0.0488***
(0.0109)
FRA 0.253*** 0.241*** 0.199*** 0.239*** 0.237*** 0.238*** 0.237*** 0.239***
(0.0163) (0.0133) (0.0385) (0.0130) (0.0127) (0.0127) (0.013) (0.014)
Treatment, cardio (t − 1) −0.00189 0.00479 0.0272 0.00557 0.0110 0.0108 0.011 0.016
(0.0322) (0.0267) (0.0880) (0.0267) (0.0255) (0.0256) (0.025) (0.027)
Treatment, muscular (t − 1) 0.0376 0.0263 −0.0165 −0.0120 0.0241 0.0246 0.025 0.038
(0.0331) (0.0261) (0.112) (0.0266) (0.0253) (0.0254) (0.025) (0.027)
Any treatment (t − 1) −0.0240 −0.0207 −0.0614 −0.0378** −0.0224 −0.0223 −0.023 −0.034*
(0.0181) (0.0145) (0.0447) (0.0146) (0.0139) (0.0139) (0.014) (0.014)
Psychotropic medication (t − 1) 0.0340* 0.0326* −0.00616 0.0470*** 0.0325* 0.0324* 0.032* 0.031*
(0.0167) (0.0135) (0.0380) (0.0132) (0.0128) (0.0128) (0.013) (0.013)
Medication purchases (t − 1)
4 to 7 −0.0113 −0.00580 0.0306 −0.00545 −0.00599 −0.00566 −0.006 −0.008
(0.0107) (0.00875) (0.0325) (0.00876) (0.00854) (0.00855) (0.009) (0.009)
8 or more 0.0188 0.0100 0.0463 0.0197* 0.0134 0.0139 0.014 0.016
(0.00987) (0.00805) (0.0293) (0.00806) (0.00784) (0.00786) (0.008) (0.008)
Sickness absenses (t − 1)
1–14 days 0.0354 0.0415* −0.0137 0.0524** 0.0337* 0.0335* 0.034* 0.042*
(0.0203) (0.0166) (0.104) (0.0172) (0.0165) (0.0165) (0.016) (0.018)
15–60 days 0.0196 0.0303 −0.0747 0.0850*** 0.0224 0.0217 0.023 0.030
(0.0210) (0.0164) (0.0585) (0.0175) (0.0159) (0.0159) (0.016) (0.017)
Over 60 days 0.0821** 0.0863*** −0.0571 0.361*** 0.0685** 0.0697** 0.070** 0.073**
(0.0315) (0.0243) (0.115) (0.0237) (0.0237) (0.0237) (0.024) (0.025)
Mean retirement rate 0.274 0.248 0.459 0.283 0.265 0.265 0.265 0.273
Observations 8376 11,862 1061 13,244 12,923 12,923 12,923 11,712
R‐squared 0.326 0.316 0.449 0.297 0.323 0.321 0.321 0.306
FE YES YES YES YES YES YES YES YES
Individual controls YES YES YES YES YES YES YES YES
Spouse controls YES YES YES YES YES YES YES YES
Health controls t − 1 YES YES YES YES YES YES YES YES
(Continues)
551 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
not necessarily be entirely innocuous. We are interested in in-
dividuals with different health statuses, and one may worry
whether excluding individuals with the poorest health affects
our results. In this section, however, we argue that the reliability
of our results is not compromised by this choice.
First, Table 6 shows that the majority of disability retirements
occur before the age of 62 and this share has remained similar
after the reform. However, an additional worry is that the re-
form might have affected the attractiveness of the different
retirement pathways. The change in the full retirement age also
affected the eligibility ages for disability pensions: after the re-
form, 63‐ to 65‐year‐old individuals could no longer retire due to
disability but were able to claim full old‐age pension. The re-
form also decreased the target age of the projected pension
component and changed the accrual rate of the projected
pension component from 0.8% to 1.3% and after 2010 it was
further increased to 1.5%. Disability pensions were calculated
with the new rules for the first time in 2006. In addition, the
reform abolished the unemployment pension and this was
partly compensated by providing occupational disability criteria
for disability retirement.
To account for the role of disability pensions, we carry out a
robustness check where we include individuals over the age of
62 who retired due to disability in our sample. This increases the
sample size by around 300 observations. However, there is a
possible endogeneity problem related to the assignment of
different retirement pathways, since the disability pensions are
higher and the incentives to postpone retirement are poorer
compared to old‐age retirement. Therefore, if we were to
calculate retirement incentives according to the rules for
disability pension, for all individuals who retired due to
disability, we would likely end up overestimating the effect of
incentives and underestimating the role of pension wealth.
To deal with this potential endogeneity issue, we follow the
example of Johansson, Laun, and Palme (2014) and use a
probabilistic approach to weight different pathways to retire-
ment. We first calculate the share of individuals aged 62 to 6521
retiring through old‐age‐retirement or disability retirement by
gender, education and year conditional on not having retired
earlier. For this purpose we extended the sample to cover all
working sectors (excluding individuals with personal retirement
ages), and the descriptive statistics on the shares are shown in
the Appendix D, Tables A4 and A5. Then these shares are used
to form a weighted sum of pension wealth:
PW(weighted) = p(DI)p(DI) + p(OLD)PW(DI)
+
p(OLD)
p(DI) + p(OLD)PW
(4)
Where p(DI) is the mean disability retirement rate and p(OLD)
is the mean old‐age retirement rate. As mentioned earlier, these
rates vary according to gender, age and year. PW is pension
wealth with old‐age pensions and PW(DI) is pension wealth
with disability pensions. This weighted pension wealth is used
to form the incentives to postpone retirement, when the possi-
bility to retire due to disability is taken into account.
The baseline estimation results are displayed in the fourth col-
umn of Table 5 and the sub‐group results are shown in the
Appendix, Table A9. Not surprisingly, the health variables are
much more important determinants of retirement when
disability retirement is taken into account. In particular, having
a large number of sickness absence days increases the likelihood
of retirement substantially (around 36 pp.). Overall the coeffi-
cient for the incentives to postpone retirement are almost the
same as in the baseline estimation. The results of the sub‐group
analysis are qualitatively similar to our baseline results, but
differences in reactions to incentives between individuals with
and without a previous spell of sickness absence are not statis-
tically significant in this analysis.
5.4 | IV Estimations
As explained in Section 3.5, we check the robustness of our
baseline results using a complementary estimation strategy
building on an IV approach.
In the first stage of the IV analysis, the dependent variable is the
change in pension wealth, Δ ln(PW), which is regressed on the
Gruber‐Saez (2002) type instrument, constructed using lagged
TABLE 5 | (Continued)
OLS 2SLS
(1) (2) (3) (4) (5) (6) (7) (8)
First stage: Dependent variable is the relative change in pension wealth calculated by lagged (one or 2 year) income
Kleibergen‐Paap F statistics 58,220 12,248
First stage r‐squared 0.227 0.223
Note: In model (1), individuals entitled to national pension are excluded, in model (2) unemployed in the previous year are excluded, in model (3) only the unemployed in
the previous year are included, in model (4) disability retirements are included and pension wealth and the incentives calculated according to Equation (4), in model (5)
pension wealth is calculated using lagged income for all, in model (6) Δ ln(pension) and ln(pension), are used to measure the incentives. OLS is used for the first 6
models, and in all models except model (6) incentives are measured using the relative change in pension wealth and the level of pension wealth. In models (7)
and (8) the estimation is conducted using 2SLS with the main sample, but year 2001 is excluded from model (8). In model (7) the relative change in pension
wealth calculated by 1 year lagged earnings  lnPWt+1  yt ,…)) − lnPWt yt−1 ,…))) are used as instruments. In model (8) the relative change in pension wealth
calculated by 2 year lagged earnings  lnPWt+1 yt−1 ,…)) − lnPWt  yt−2 ,…))) are used as instruments. Δ ln(PW) is the financial incentives to postpone retirement
(i.e. the relative change in pension wealth, PW , when retirement is postponed by 1 year). FRA is an indicator for reaching full retirement age. Sickness absences
indicates days after the waiting period. Robust standard errors in parentheses ***p < 0.001, **p < 0.01, *p < 0.05.
552 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
income but current year's policy rules. The results regarding the
mean impacts are shown in Columns 7 (using the instrument
with 1 year lagged incomes) and 8 (2 year lagged incomes) of
Table 5. The first‐stage F‐test result is very strong, confirming
instrument strength. The results are qualitatively very similar to
the OLS results ‐ the signs of the main coefficients of interest are
as expected and the estimated coefficients are also statistically
highly significant. This indicates that the potential endogeneity
in the incentive measure did not turn out to plague our baseline
OLS analysis. This increases our confidence that we have indeed
isolated a causal relationship between the financial incentives to
postpone retirement and the retirement decision.
Tables A10 and A11 show the 2SLS estimation results for the
different population groups. Qualitatively, the picture that
emerges is similar to the main analysis. Again, the differences in
the reactions across population groups are typically not statis-
tically significant. Only individuals with a sickness absence spell
seem to differ in their reactions as they react less than those
without a sickness absence spell. The sub‐group IV results are,
therefore, similar to the corresponding OLS results.
5.5 | Additional Robustness Checks
In columns (5) and (6) of Table 5 and in Appendix B, we present
two additional robustness checks, dealing with changes in
measuring incentives andusing interaction terms, instead of split‐
sample analysis, to examineheterogeneity in responses. Choosing
to use the annual change in pension, rather than pension wealth,
as an incentive measure leads to very similar results. When using
interaction terms, the only statistically significant difference in
the response pertains to sick leave, further confirming the results
discussed above.
6 | Conclusions
We analyzed the effects of retirement incentives on retirement
behavior, using the Finnish pension reform implemented in
2005. The reform changed the financial incentives to postpone
retirement differently for people with different labor market
histories, and we exploit the exogenous variation in retirement
incentives generated by the reform. The results indicate that the
changes in financial incentives to postpone retirement influ-
enced retirement decisions in an expected manner on average.
Improved incentives to continue working induced individuals to
postpone retirement, while a higher level of accrued pension
wealth led to earlier retirement.
Our second key aim was to study how the reactions to incentives
vary among different types of people. In particular, we analyzed
how health status, measured using rich administrative data,
may modify the effect of financial incentives on actual retire-
ment decisions: Are all individuals, also those with ill health,
able to take advantage of the monetary incentives to postpone
retirement? On the one hand, older workers who have health
problems may find it difficult to postpone retirement despite the
financial gain of working longer. On the other hand, the
financial gain may be especially important for those with worse
health, as greater financial compensation is needed to offset the
(mental or physical) costs of working. Depending on which
mechanism dominates, ill health may lead to a muted or a more
pronounced reaction to changes in incentives. To the best of our
knowledge, similar analysis combining quasi‐experimental
variation in retirement incentives, and focusing on this type of
potential heterogeneity in the responses to these incentives, has
not been offered in previous literature. While we implement this
research in the Finnish context, we argue that the results have
external validity, for example because of the objective health
measures used, and because the types of incentives we examine
TABLE 6 | Share of retirements by age and retirement type.
Age 2003 2004 2005 2006 2007 2008
Old‐age retirements
62 5% 4% 7% 9% 13% 13%
63 15% 17% 35% 36% 42% 44%
64 2% 5% 17% 11% 10% 10%
65 44% 44% 24% 23% 17% 18%
66 1% 1% 1% 1% 1% 2%
67 0% 0% 0% 0% 1% 1%
68 0% 0% 0% 0% 0% 1%
Total number of retirements 19,768 21,909 31,331 27,733 28,772 35,768
Retirement due to disability
62 4% 3% 3% 3% 3% 4%
63 2% 2% 2% 1% 1% 1%
64 1% 1% 1% 0% 0% 0%
65 0% 0% 0% 0% 0% 0%
Total number of retirements 28,056 28,386 28,746 27,215 28,187 28,665
Note: Values include all working sectors. Shares are calculated relative to the total number of retirements of each type and year. Age is at the end of the year.
Source: Own calculations based on Finnish Centre for Pensions online database.
553 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
are increasingly typical in pension reforms implemented in
high‐income countries.
Using a wide array of health indicators, including inpatient
hospital care and purchases of prescribed medication, we found
that individuals with various different types of health issues do
respond to incentives to postpone retirement. There are reasons
to interpret these findings with some caution, as the health
measures used might still hide some heterogeneity between
individuals in the severity of morbidity.
An interesting exception to the general pattern highlighted
above was that those with a sickness absence period exceeding
10 days—a measure of ill health specifically related to the ability
to work—reacted less strongly than others to financial in-
centives to continue working. The individuals in this group also
have many more health issues as indicated by the other health
measures used, which suggests that the group has particularly
severe conditions affecting functional ability. Sickness absen-
teeism further suggests that prior treatment of these conditions
has not been effective in maintaining the individuals work ca-
pacity. The statistical significance of these results varies some-
what, however, depending on the sample used.
Overall, our results suggest that many different types of in-
dividuals, who receive treatment for different health conditions,
are able to take advantage of financial incentives inherent in the
pension system. However, individuals whose health problems
have led to reduced working capabilities may not respond to
financial incentives to postpone retirement in the same way as
their more healthy peers. The associated equity/efficiency trade‐
off may be mitigated by ensuring that there are suitable mech-
anisms in place, via for example disability insurance, for those
who truly cannot continue working.
Acknowledgments
This research was financially supported by NORFACE (Grant No. 462–
14‐013), Yrjö Jahnsson foundation, the INVEST Research Flagship
(Research Council of Finland decision number: 345546), the Finnish
Centre of Excellence in Tax Systems Research (FIT) (Research Council
of Finland decision numbers: 346250 and 346251), the European
Research Council under the European Union's Horizon 2020 research
and innovation programme (grant agreement No 101019329), the Stra-
tegic Research Council (SRC) within the Research Council of Finland
grants for ACElife (\#352543–352572) and LIFECON (\#345219), and
grants to the Max Planck—University of Helsinki Center from the Jane
and Aatos Erkko Foundation, the Max Planck Society, University of
Helsinki, and Cities of Helsinki, Vantaa and Espoo. The study does not
necessarily reflect the Commission's views and in no way anticipates the
Commission's future policy in this area. The funders had no role in the
study design, data collection and analysis, decision to publish, or
preparation of the manuscript. We benefited from helpful comments by
the referees, and we are also grateful to participants at 75th Annual
Congress of the IIPF, 42nd Annual Meeting of the Finnish Economic
Association, as well as seminar participants at the Finnish Institute for
Health and Welfare, Labour Institute for Economic Research, Finnish
Centre for Pensions and University of Turku, for their comments and
discussions.
Conflicts of Interest
The authors declare no conflicts of interest.
Data Availability Statement
Due to data protection regulations of the national register‐holders
providing the data, we are not allowed to make the data available to
third parties. Interested researchers have the possibility to obtain data
access by contacting the following register‐holding public institutions:
1. Statistics Finland (http://www.stat.fi/tup/mikroaineistot/index_
en.html). Contact by email tutkijapalvelut@stat.fi.
2. Findata (https://findata.fi/en/). Contact by email info(at)findata.fi.
Findata handles applications for health data provided by the
Finnish Institute for Health and Welfare, and the Social Insurance
Institution.
3. Finnish Centre for Pensions (https://www.etk.fi/en/).
Endnotes
1Gustman and Steinmeier (2018) also examine how health influences
the role of economic incentives for the decision to retire and found that
health does not affect responses to retirement incentives. However,
their approach is very different to ours. They build a structural labour
supply model with retirement and then simulate the impacts of
changing incentives on people retiring, whereas we use a quasi‐
experimental setting to obtain causal estimates of how incentives in-
fluence retirement decisions among people with different health status.
2 In a somewhat different context, Hall et al. (2022) analyzed individual
differences in reactions to an active labour market programme tar-
geted at young people, and found that young individuals who are in a
difficult overall life situation (e.g. school dropouts and individuals
suffering from mental health problems) do not react to activation.
This implies that active labour market policies may not help in-
dividuals at risk of social exclusion. Since health problems are more
relevant for the elderly, studying related issues in the context of
pension reforms appears particularly relevant.
3 50% consumer price index and 50% earnings index.
4 80% consumer price index and 20% earnings index.
5 20% consumer price index and 80% earnings index.
6 The reform did not change the eligibility age for full national pension
or the penalty for early claiming within that system.
7Other measures for the incentive to stay in the labour force used in the
literature are the option value measure (see Stock and Wise (1990))
and the peak value of pension wealth (see Coile and Gruber (2007)).
The option value is formed by calculating the individual's utility at
different retirement dates. These levels of utilities are compared to that
with the highest utility (optimal retirement date) and this difference
gives the measure for the incentive to retire at any given point in time.
We choose not to use this measure, because it would require making
additional assumptions about individuals' risk aversion and con-
sumption and leisure preferences. The peak value is similar to the
option value, without the need for making additional assumptions
about individual preferences. The peak value approach compares the
highest level of pension wealth with the current‐year level of pension
wealth. One potential worry with the peak value measure is that it
does not take into account how many years are needed to postpone
retirement to reach the optimal level of pension wealth.
8 Before 2005, pensionswere tied to the halfway index (for under 65‐year‐
olds) and to the earnings‐related pension index (those 65 or older) and
after that only to the earnings‐related pension index. National pensions
are tied to the national pension index (before and after the reform).
9 For the first one we use 1.59%, for the second one we use 3.13% and
for the third one 1.8%. We also convert pension wealth into year 2000
euros using the consumer price index.
10 Appendix: The Effect of the Reform on Incentives According to
Health Statuses (Figures A2 and A3) contains some descriptive figures
with earnings‐related pensions only.
554 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
11 In Appendix C (Figures A4–A13) we show how the incentives and the
effect of the reform on incentives vary according to health status. The
key takeaway from these charts is that the changes in incentives are
very similar across individuals with different health status, and
therefore it is unlikely that any difference in reactions to incentives
between healthy and unhealthy individuals would stem from facing
different incentives.
12We identify hospitalisations and treatments due to diseases of the
circulatory system (I00‐I99) and diseases of the musculoskeletal sys-
tem and connective tissue (M00‐M99).
13Medication with Anatomical Therapeutic Chemical (ATC) classifica-
tion codes relating to psycholeptics (N05) and psychoanaleptics (N06,
excluding anti‐dementia drugs N06D) were classified as psychotropic
medication in the data.
14 Sickness absences that continue into the next year are allocated to
both years according to the respective number of days.
15 These data do not cover information on occupational healthcare.
However, this is unlikely to be a major limitation, given that the role
of occupational healthcare in specialized care is non‐existent.
16 A full description of the control variables used in the estimations is
presented in Appendix A.
17We use also another instrument were the earnings are lagged by
2 years, formally lnPWt+1 yt−1,…)) − lnPWt  yt−2,…)). Longer lags are
more likely to be unaffected by possible mean reversion issues
(Weber 2014): They are unrelated to the error term when there is first‐
order autocorrelation.
18 Before the reform, reaching full retirement age was assigned value
one when an individual turned 65. After the reform, it was assigned
value one when an individual turned 63. In our analysis, year 2005 is
excluded, so after the reform, only individuals turning 63 receive
value one for the FRA variable. (In year 2005, individuals turning 63,
64, and 65 would have received value one for the FRA variable.)
19 On the other hand, the estimated coefficient for any treatment period,
a treatment period due to musculoskeletal issues, a treatment period
due to cardiovascular diseases, less than 8 medical purchases and
having a length of sickness absences of 15–60 days are not statistically
significant.
20 Both employed and unemployed persons are entitled to sickness
allowance in Finland, but because the level of compensation is the
same in both benefits, the unemployed may not be motivated to apply
for sickness allowance.
21 After the reform only those aged 62 to 63 are included.
22 Figure A1 includes national pensions as well. National pensions partly
offset the rise in pensionwealth, since the statutory ages for full national
pension did not change. The 63‐year‐olds who are not eligible for a
national pension sawan increase of around 10% in their pensionwealth.
References
Bernheim,B.D.,D. Ray, and Ş.Yeltekin. 2015. “Poverty and Self‐Control.”
Econometrica 83, no. 5: 1877–1911. https://doi.org/10.3982/ecta11374.
Brown, K. M. 2013. “The Link Between Pensions and Retirement
Timing: Lessons From California Teachers.” Journal of Public Eco-
nomics 98: 1–14. https://doi.org/10.1016/j.jpubeco.2012.10.007.
Carvalho, L. S., S. Meier, and S. W. Wang. 2016. “Poverty and Economic
Decision‐Making: Evidence From Changes in Financial Resources at
Payday.” American Economic Review 106, no. 2: 260–284. https://doi.
org/10.1257/aer.20140481.
Coile, C., and J. Gruber. 2007. “Future Social Security Entitlements and
the Retirement Decision.” Review of Economics and Statistics 89, no. 2:
234–246. https://doi.org/10.1162/rest.89.2.234.
Decker, S., and H. Schmitz. 2016. “Health Shocks and Risk Aversion.”
Journal of Health Economics 50: 156–170. https://doi.org/10.1016/j.
jhealeco.2016.09.006.
Dohmen, T., A. Falk,D.Huffman, andU. Sunde. 2010. “AreRiskAversion
and ImpatienceRelated to CognitiveAbility?”AmericanEconomic Review
100, no. 3: 1238–1260. https://doi.org/10.1257/aer.100.3.1238.
Engels, B., J. Geyer, and P. Haan. 2017. “Pension Incentives and Early
Retirement.” Labour Economics 47: 216–231. https://doi.org/10.1016/j.
labeco.2017.05.006.
Furgeson, J., R. P. Strauss, and W. B. Vogt. 2006. “The Effects of Defined
Benefit Pension Incentives and Working Conditions on Teacher
Retirement Decisions.” Education Finance and Policy 1, no. 3: 316–348.
https://doi.org/10.1162/edfp.2006.1.3.316.
Garcia‐Gomez, P., T. J. Galama, E. van Doorslaer, and A. Lopez Nicolas.
2017. “Interactions Between Financial Incentives and Health in the Early
Retirement Decision.” CESR‐Schaeffer Working Paper No. 2017‐004.
https://doi.org/10.2139/ssrn.2971285.
Gloede, O., L. Menkhoff, and H. Waibel. 2015. “Shocks, Individual Risk
Attitude, and Vulnerability to Poverty Among Rural Households in
Thailand and Vietnam.” World Development 71: 54–78. https://doi.org/
10.1016/j.worlddev.2013.11.005.
Gruber, J., O. Kanninen, and T. Ravaska. 2022. “Relabeling, Retirement
and Regret.” Journal of Public Economics 211: 104677. https://doi.org/
10.1016/j.jpubeco.2022.104677.
Gruber, J., and E. Saez. 2002. “The Elasticity of Taxable Income: Evi-
dence and Implications.” Journal of Public Economics 84, no. 1: 1–32.
https://doi.org/10.1016/S0047‐2727(01)00085‐8.
Gustman, A. L., and T. L. Steinmeier. 2018. “The Role of Health in
Retirement.” In Transitions through the Labor Market (Research in Labor
Economics), Vol. 46, 229–297: Chapter 7. Leeds: Emerald Publishing
Limited. https://doi.org/10.1108/S0147‐912120180000046007.
Hall, C., K. Kotakorpi, L. Liljeberg, and J. Pirttilä. 2022. “Screening
Through Activation? Differential Effects of a Youth Activation Pro-
gram.” Journal of Human Resources: 1117–9182R1.
Hanel, B. 2010. “Financial Incentives to Postpone Retirement and
Further Effects on Employment – Evidence From a Natural Experi-
ment.” Labour Economics 17, no. 3: 474–486. https://doi.org/10.1016/j.
labeco.2009.10.001.
Hernæs, E., S. Markussen, J. Piggott, and K. Røed. 2016. “Pension Reform
and Labor Supply.” Journal of Public Economics 142: 39–55. https://doi.
org/10.1016/j.jpubeco.2016.08.009.
Johansson, P., L. Laun, and M. Palme. 2014. “Pathways to Retirement
and the Role of Financial Incentives in Sweden.” Working Paper Series
2014:20, IFAU ‐ Institute for Evaluation of Labour Market and Education
Policy.
Kerkhofs, M., M. Lindeboom, and J. Theeuwes. 1999. “Retirement,
Financial Incentives and Health.” Labour Economics 6, no. 2: 203–227.
https://doi.org/10.1016/S0927‐5371(99)00016‐0.
Leijten, F. R. M., A. de Wind, S. G. van den Heuvel, et al. 2015. “The
Influence of Chronic Health Problems and Work‐Related Factors on
Loss of Paid Employment Among Older Workers.” Journal of Epide-
miology & Community Health 69, no. 11: 1058–1065. https://doi.org/10.
1136/jech‐2015‐205719.
Leinonen, T., M. Laaksonen, T. Chandola, and P. Martikainen. 2016.
“Health as a Predictor of Early Retirement before and After Introduction
of a Flexible Statutory Pension Age in Finland.” Social Science & Med-
icine 158: 149–157. https://doi.org/10.1016/j.socscimed.2016.04.029.
Mani, A., S. Mullainathan, E. Shafir, and J. Zhao. 2013. “Poverty Impedes
Cognitive Function.” Science 341, no. 6149: 976–980. https://doi.org/10.
1126/science.1238041.
555 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Manoli, D., and A. Weber. 2016. “Nonparametric Evidence on the Effects
of Financial Incentives on Retirement Decisions.” American Economic
Journal: Economic Policy 8, no. 4: 160–182. https://doi.org/10.1257/pol.
20140209.
OECD. 2019. “Pensions at a Glance 2019.”
Shah, A. K., S. Mullainathan, and E. Shafir. 2012. “Some Consequences
of Having Too Little.” Science 338, no. 6107: 682–685. https://doi.org/10.
1126/science.1222426.
Staubli, S., and J. Zweimüller. 2013. “Does Raising the Early Retirement
Age Increase Employment of Older Workers?” Journal of Public Eco-
nomics 108: 17–32. https://doi.org/10.1016/j.jpubeco.2013.09.003.
Stock, J. H., and D. A. Wise. 1990. “Pensions, the Option Value of Work,
and Retirement.” Econometrica 58, no. 5: 1151–1180. https://doi.org/10.
2307/2938304.
Sund, R. 2012. “Quality of the Finnish Hospital Discharge Register: A
Systematic Review.” Scandinavian Journal of Public Health 40, no. 6:
505–515. https://doi.org/10.1177/1403494812456637.
Tanaka, T., C. F. Camerer, and Q. Nguyen. 2010. “Risk and Time
Preferences: Linking Experimental and Household Survey Data From
Vietnam.” American Economic Review 100, no. 1: 557–571. https://doi.
org/10.1257/aer.100.1.557.
Uusitalo, R., and S. Nivalainen. 2013. Vuoden 2005 Eläkeuudistuksen
Vaikutus Eläkkeellesiirtymisikään. Prime Minister’s Office. Valtioneu-
voston kanslian raporttisarja 5/2013. http://urn.fi/URN:NBN:fi‐
fe201704066002.
van Rijn, R. M., S. J. W. Robroek, S. Brouwer, and A. Burdorf. 2014.
“Influence of Poor Health on Exit From Paid Employment: A Systematic
Review.” Occupational and Environmental Medicine 71, no. 4: 295–301.
https://doi.org/10.1136/oemed‐2013‐101591.
Weber, C. 2014. “TowardObtaining aConsistent Estimate of the Elasticity
of Taxable Income Using Difference‐In‐Differences.” Journal of Public
Economics 117: 90–103. https://doi.org/10.1016/j.jpubeco.2014.05.004.
Appendix A: Variables
The full list of control variables used in the estimations is presented
here.
Individual controls: Gender, age at the end of the year, working history
(≤ 35 years, 35–40 years, at least 40 years), level of education, field of
education, region, reached full retirement age (yes or no), reached early
retirement age (yes or no) and language. Spouse controls: Spouse (yes or
no), age of spouse over early retirement age (yes or no) and age of
spouse over full retirement age (yes or no). Health controls: prescriptive
drug purchases (0–3, 4 to 7 and 8 or more), purchase of psychotropic
medication (yes or no), any treatment period (yes or no), treatment
period for circulatory or musculoskeletal diseases (yes or no), sickness
absences (1–14 days, 15–60 days and over 60 days).
Appendix B: Further Robustness Checks
Additional Description of the Changes in Incentives
Figure A1 shows the direct effect of the reform, in percentages, on
pension wealth and incentives for those aged 60 to 66. The direct effect
on pension wealth is calculated using the 2004 data and the information
on total accrued pensions at the end of year 2004. As can be seen from
the figure, people below the age of 65 saw an overnight increase in their
pension wealth, since the eligibility age for old‐age pension decreased
from 65 to 63 years of age. The effect was the highest for 63‐year‐old
individuals, who saw an increase of around 9% in their pension
wealth.22 The pensions of individuals who were at least 65 years old at
the time of the reform were protected. Therefore, the change in pension
wealth is zero in the figure for those age groups.
In addition to the overnight effects shown in Figure A1, the reform also
had other effects on pension wealth. The larger accrual rate for those
aged 53–60 years or at least 63 years, the change in the indexation of
accrued pensions and the new feature that pensions accruing after the
age of 63 did not affect national pensions all contributed to increase
pension wealth. On the other hand, the smaller accrual rate for those
aged 60–62 years, the change in delayed claiming, and the introduction
of the life‐expectancy coefficient had a decreasing effect on pension
wealth.
The log‐difference of pension wealth (incentives to postpone retirement)
changed as well. The reform increased the incentives for those under
62.5 years, weakened incentives for those between 62.5 and 65 years,
while the change in incentives was close to zero for those over 65 years
of age.
Robustness: Different Incentive Measures
In our baseline analysis, pension wealth was calculated using current
earnings for those who were not retired or retiring in the current year.
In column 5 of Table 5 we have tested how the results change if the
pension wealth is calculated with 1 year lagged income for all in-
dividuals. Again the results are nearly identical with our main analysis
indicating that the choice of income reference year does not drive the
results. Similarly, in the subgroup analysis, the results closely resemble
those of our main analysis (shown in Table A8).
In our main analysis we formed the measure for the incentives using
changes in pension wealth, which is the discounted stream of future
pensions until age 100. One possible issue with our choice is that the
time horizon is too long. To test for the robustness of our results against
this choice, we run the analysis using the other extreme, that is using
annual pension instead of pension wealth, and using the accrual rate of
the pension as the measure for the incentives to postpone retirement.
The results are shown in column 6 of Table 5 and they are very similar
to the results of our baseline analysis. For the subgroup analysis, the
results are also very similar (shown in Table A1).
Sub‐Group Analysis Using Interaction Terms
In our primary analysis, we examined the variations in reactions to
incentives across different health statuses by dividing the sample into
distinct subgroups. In this section, we further evaluate the robustness of
our findings by introducing interaction terms involving incentives,
pension wealth, reaching full retirement age, and health variables.
Formally, we estimate the following equation using OLS:
Ri,t = θ1Δ lnPWi,t) + θ2Δ lnPWi,t) ×Healthi,t−1 + θ3 lnPWi,t)
+ θ4 lnPWi,t) ×Healthi,t−1 + θ5FRA + θ6FRA ×Healthi,t−1
+ θ7Healthi,t−1 + γt + βXi,t + ui,t (A1)
where Healthi,t−1 is an indicator variable for having ill health (in
the same domains as earlier) for individual i and all the other
variables are defined in the same manner as in Equation (3). For
this analysis, we employ the same set of health indicators as
used in our primary subgroup analysis.
Table A2 presents the results of our subgroup analysis with interaction
terms. The first three rows illustrate the impact of incentives, pension
wealth, and reaching full retirement age on the retirement risk for
individuals without the specific health status. Subsequently, the
following three interactions describe how much the impact of these
variables differs for individuals with the particular health status.
Therefore, the total effect of these variables for individuals with the
specific health status is the sum of these two effects. The last row
demonstrates the impact of the specific health issue on the risk of
retirement.
The findings in Table A2 support our earlier results. According to the
interaction terms, the only statistically significant difference emerges
among individuals with a sickness absence spell in the previous year.
556 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
This group appears to be less responsive to incentives compared to
those without such absence spells. The results indicate that a 1% point
increase in incentives reduces the risk of retirement by approximately
1.4% points (calculated as −2.326 þ 0.893) for individuals with sickness
absence spells, while the estimated effect remains at −2.326 for those
without such absence spells.
FIGURE A1 | Changes in pension wealth and incentives to postpone retirement in the 2005 reform. The values are calculated by comparing the
levels of pension wealth using the retirement rules before and after the reform. These are calculated as averages within bimonthly bins. The red line is
calculated as a percentage point difference in the relative change in pension wealth when retirement is postponed by 12 months that is the line depicts
the change in the financial incentive to postpone retirement, as defined in Equation (2). The blue line is calculated as a percentage change in pension
wealth. The change in pension wealth is calculated using the information on total accrued pension at the end of year 2004 and using the 2004 data.
Incentives are calculated using the whole sample. Age is age at the end of the year. [Colour figure can be viewed at wileyonlinelibrary.com]
557 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Appendix C: Additional Figures
The Effect of the Reform on Incentives Without National
Pensions
FIGURE A2 | The distribution of the change in retirement incentives without national pensions (EUR) in the 2005 reform. The figure displays the
distribution of the change implied by the 2005 reform in the financial incentive to postpone retirement, measured in euros. Pension wealth includes
only earnings‐related pensions. The values are in year 2000 euros and extreme values are excluded. [Colour figure can be viewed at
wileyonlinelibrary.com]
FIGURE A3 | The distribution of the change in retirement incentives without national pensions (EUR) in the 2005 reform by age groups. The
figure displays the distribution of the change implied by the 2005 reform in the financial incentive to postpone retirement, measured in euros.
Pension wealth includes only earnings‐related pensions. The values are in year 2000 euros and extreme values are excluded. [Colour figure can
be viewed at wileyonlinelibrary.com]
558 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
The Incentives to Postpone Retirement According to Health
Statuses
FIGURE A4 | The distribution of retirement incentives (EUR) by ill health. The figure displays the distribution of the change in the amount of
pension wealth when retirement is postponed by 1 year that is the financial incentive to postpone retirement, measured in euros. Pension wealth
includes both earnings‐related pensions and national pensions. The values are in year 2000 euros and extreme values are excluded. [Colour figure
can be viewed at wileyonlinelibrary.com]
FIGURE A5 | The distribution of retirement incentives (EUR) by medication purchases. The figure displays the distribution of the change in the
amount of pension wealth when retirement is postponed by 1 year that is the financial incentive to postpone retirement, measured in euros. Pension
wealth includes both earnings‐related pensions and national pensions. The values are in year 2000 euros and extreme values are excluded. [Colour
figure can be viewed at wileyonlinelibrary.com]
559 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
FIGURE A6 | The distribution of retirement incentives (EUR) by psychotropic medication. The figure displays the distribution of the change in the
amount of pension wealth when retirement is postponed by 1 year that is the financial incentive to postpone retirement, measured in euros. Pension
wealth includes both earnings‐related pensions and national pensions. The values are in year 2000 euros and extreme values are excluded. [Colour
figure can be viewed at wileyonlinelibrary.com]
FIGURE A7 | The distribution of retirement incentives (EUR) by any treatment. The figure displays the distribution of the change in the amount
of pension wealth when retirement is postponed by 1 year that is the financial incentive to postpone retirement, measured in euros. Pension wealth
includes both earnings‐related pensions and national pensions. The values are in year 2000 euros and extreme values are excluded. [Colour figure can
be viewed at wileyonlinelibrary.com]
560 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
The Effect of the Reform on Incentives According to Health Statuses
FIGURE A8 | The distribution of retirement incentives (EUR) by any sickness absences. The figure displays the distribution of the change in the
amount of pension wealth when retirement is postponed by 1 year that is the financial incentive to postpone retirement, measured in euros. Pension
wealth includes both earnings‐related pensions and national pensions. The values are in year 2000 euros and extreme values are excluded. [Colour
figure can be viewed at wileyonlinelibrary.com]
FIGURE A9 | The distribution of the change in retirement incentives (EUR) in the 2005 reform, by ill health. The figure displays the distribution of
the change implied by the 2005 reform in the financial incentive to postpone retirement, measured in euros. Pension wealth includes both earnings‐
related pensions and national pensions. The values are in year 2000 euros and extreme values are excluded. [Colour figure can be viewed at
wileyonlinelibrary.com]
561 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
FIGURE A10 | The distribution of the change in retirement incentives (EUR) in the 2005 reform, by medication purchases. The figure displays the
distribution of the change implied by the 2005 reform in the financial incentive to postpone retirement, measured in euros. Pension wealth includes
both earnings‐related pensions and national pensions. The values are in year 2000 euros and extreme values are excluded. [Colour figure can be
viewed at wileyonlinelibrary.com]
FIGURE A11 | The distribution of the change in retirement incentives (EUR) in the 2005 reform, by psychotropic medication. The figure displays
the distribution of the change implied by the 2005 reform in the financial incentive to postpone retirement, measured in euros. Pension wealth
includes both earnings‐related pensions and national pensions. The values are in year 2000 euros and extreme values are excluded. [Colour figure
can be viewed at wileyonlinelibrary.com]
562 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
FIGURE A12 | The distribution of the change in retirement incentives (EUR) in the 2005 reform, by any treatment. The figure displays the
distribution of the change implied by the 2005 reform in the financial incentive to postpone retirement, measured in euros. Pension wealth
includes both earnings‐related pensions and national pensions. The values are in year 2000 euros and extreme values are excluded. [Colour figure
can be viewed at wileyonlinelibrary.com]
FIGURE A13 | The distribution of the change in retirement incentives (EUR) in the 2005 reform, by sickness absences. The figure displays the
distribution of the change implied by the 2005 reform in the financial incentive to postpone retirement, measured in euros. Pension wealth
includes both earnings‐related pensions and national pensions. The values are in year 2000 euros and extreme values are excluded. [Colour figure
can be viewed at wileyonlinelibrary.com]
563 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TA
BL
E
A
1
|
O
LS
es
tim
at
es
by
po
pu
la
tio
n
gr
ou
ps
w
ith
pe
ns
io
n
an
d
ac
cr
ua
lr
at
e
of
pe
ns
io
n.
Tr
ea
tm
en
t,
ca
rd
io
(t
−
1)
Tr
ea
tm
en
t,
m
us
cu
la
r
(t
−
1)
A
ny
tr
ea
tm
en
t(
t−
1)
M
ed
ic
at
io
n
(t
−
1)
Ps
yc
ho
tr
op
ic
m
ed
ic
at
io
n
(t
−
1)
Si
ck
ne
ss
ab
se
nc
es
(t
−
1)
Ill
he
al
th
(t
−
1)
Ye
s
N
o
Ye
s
N
o
Ye
s
N
o
8≤
7
or
le
ss
Ye
s
N
o
1≤
N
o
Ye
s
N
o
D
ep
en
de
nt
va
ria
bl
e
is
re
tir
em
en
td
ec
isi
on
Δ
ln
(p
en
sio
n)
−3
.1
16
**
−2
.2
06
**
*
−2
.8
53
**
*
−2
.2
19
**
*
−2
.3
64
**
*
−2
.2
23
**
*
−2
.1
53
**
*
−2
.2
65
**
*
−2
.3
44
**
*
−2
.2
16
**
*
−1
.1
05
**
−2
.3
51
**
*
−2
.2
93
**
*
−2
.1
88
**
*
(1
.0
01
)
(0
.1
27
)
(0
.8
52
)
(0
.1
28
)
(0
.3
66
)
(0
.1
34
)
(0
.2
08
)
(0
.1
59
)
(0
.3
98
)
(0
.1
33
)
(0
.4
18
)
(0
.1
32
)
(0
.1
95
)
(0
.1
66
)
ln
(p
en
sio
n)
−0
.1
25
0.
05
14
**
*
0.
05
23
0.
04
93
**
*
0.
02
62
0.
05
24
**
*
0.
05
35
**
0.
04
35
**
0.
06
24
0.
04
79
**
*
0.
04
31
0.
05
01
**
*
0.
05
23
**
0.
04
41
**
(0
.0
80
8)
(0
.0
11
0)
(0
.0
94
2)
(0
.0
10
9)
(0
.0
34
4)
(0
.0
11
5)
(0
.0
18
4)
(0
.0
13
6)
(0
.0
38
0)
(0
.0
11
4)
(0
.0
35
1)
(0
.0
11
5)
(0
.0
17
6)
(0
.0
14
0)
FR
A
0.
25
3*
0.
23
7*
**
0.
04
54
0.
24
1*
**
0.
18
5*
**
0.
24
2*
**
0.
25
4*
**
0.
22
7*
**
0.
21
9*
**
0.
23
7*
**
0.
15
6*
**
0.
24
8*
**
0.
25
3*
**
0.
22
7*
**
(0
.1
01
)
(0
.0
12
9)
(0
.1
14
)
(0
.0
12
8)
(0
.0
42
3)
(0
.0
13
4)
(0
.0
21
9)
(0
.0
15
7)
(0
.0
47
6)
(0
.0
13
3)
(0
.0
37
8)
(0
.0
13
5)
(0
.0
20
6)
(0
.0
16
2)
p‐
va
lu
e
Δ
ln
(p
en
sio
n)
0.
31
6
0.
41
8
0.
71
3
0.
66
7
0.
75
6
0.
00
4
0.
68
p‐
va
lu
e
ln
(p
en
sio
n)
0.
01
6
0.
97
2
0.
46
3
0.
65
8
0.
70
8
0.
84
7
0.
71
2
p‐
va
lu
e
FR
A
0.
86
1
0.
06
0.
19
1
0.
30
3
0.
70
6
0.
02
0.
30
7
O
bs
er
va
tio
ns
28
7
12
,6
36
31
3
12
,6
10
14
47
11
,4
76
45
01
84
22
10
88
11
,8
35
15
43
11
,3
80
49
90
79
33
R‐
sq
ua
re
d
0.
43
8
0.
31
9
0.
41
2
0.
32
0
0.
32
0
0.
32
2
0.
31
2
0.
32
7
0.
31
2
0.
32
4
0.
34
3
0.
32
0
0.
31
3
0.
32
7
FE
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
In
di
vi
du
al
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
Sp
ou
se
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
H
ea
lth
co
nt
ro
ls
t−
1
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
ot
e:
Th
e
nu
ll
hy
po
th
es
is
fo
rt
he
p‐
va
lu
es
is
th
at
th
e
di
ffe
re
nc
e
be
tw
ee
n
co
ef
fic
ie
nt
se
qu
al
sz
er
o.
Δ
ln
(p
en
sio
n)
is
th
e
re
la
tiv
e
ch
an
ge
in
pe
ns
io
n
w
he
n
re
tir
em
en
ti
sp
os
tp
on
ed
by
1
ye
ar
.S
ic
kn
es
sa
bs
en
ce
si
nd
ic
at
es
da
ys
af
te
r
th
e
w
ai
tin
g
pe
rio
d.
Ro
bu
st
sta
nd
ar
d
er
ro
rs
in
pa
re
nt
he
se
s
**
*p
<
0.
00
1,
**
p
<
0.
01
,*
p
<
0.
05
.
A
pp
en
di
x
D
:A
dd
iti
on
al
Ta
bl
es
564 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TABLE A2 | Sub‐group analysis (OLS) utilizing interaction terms.
Treatment,
cardio (t − 1)
Treatment,
muscular (t − 1)
Any
treatment
(t − 1)
Medication
(t − 1)
(8 or more)
Psychotropic Sickness Ill
health
(t − 1)
Medication
(t − 1)
Absences
(t − 1)
Dependent variable is retirement decision
Δ ln(PW ) −2.213*** −2.231*** −2.236*** −2.286*** −2.223*** −2.326*** −2.213***
(0.127) (0.127) (0.133) (0.153) (0.132) (0.131) (0.158)
ln(PW ) 0.0493*** 0.0463*** 0.0482*** 0.0401*** 0.0461*** 0.0461*** 0.0396**
(0.0109) (0.0109) (0.0111) (0.0121) (0.0110) (0.0111) (0.0123)
FRA 0.237*** 0.237*** 0.237*** 0.227*** 0.238*** 0.239*** 0.228***
(0.0128) (0.0128) (0.0131) (0.0142) (0.0130) (0.0131) (0.0145)
Δ ln(PW) ×
Healtht−1
−0.776 0.110 0.0883 0.177 −0.0213 0.893* −0.0133
(0.784) (0.748) (0.344) (0.233) (0.371) (0.379) (0.228)
ln(PW) ×
Healtht−1
−0.0575 0.0986 0.000111 0.0212 0.0269 0.0278 0.0204
(0.0493) (0.0557) (0.0240) (0.0154) (0.0276) (0.0263) (0.0151)
FRA
× Healtht−1
−0.0184 −0.0144 0.000612 0.0302 −0.00885 −0.0194 0.0241
(0.0568) (0.0580) (0.0276) (0.0182) (0.0313) (0.0275) (0.0178)
Healtht−1 0.797 −1.205 0.000491 −0.264 −0.293 −0.371 −0.238
(0.613) (0.692) (0.297) (0.190) (0.341) (0.324) (0.186)
Observations 12,923 12,923 12,923 12,923 12,923 12,923 12,923
R‐squared 0.319 0.320 0.319 0.320 0.320 0.320 0.320
FE YES YES YES YES YES YES YES
Individual
controls
YES YES YES YES YES YES YES
Spouse controls YES YES YES YES YES YES YES
Health
controls t − 1
NO NO NO NO NO NO NO
Note: Healtht−1 is an indicator variable representing various health statuses, with its interpretation varying across each model. Δ ln(PW) is the financial
incentives to postpone retirement (i.e. the relative change in pension wealth, PW , when retirement is postponed by 1 year). FRA is an indicator for reaching full
retirement age. The interaction terms, Δ ln(PW)× Healtht−1, ln(PW)× Healtht−1, and FRA × Healtht−1, capture the interplay between incentives, pension
wealth, reaching full retirement age, and health variables. Interaction terms depict the situation where health indicator gets the value 1. In interaction term,
FRA × Healtht−1, FRA also gets the value 1. Sickness absences indicates days after the waiting period. Robust standard errors in parentheses ***p < 0.001,
**p < 0.01, *p < 0.05.
TABLE A3 | Descriptive statistics by sickness absences.
Sickness absences (t − 1)
No 1 ≤
Share 0.881 0.119
Retirement rate 0.286 0.262
Reaching full retirement age 0.265 0.255
Female 0.459 0.399
Age (at the end of the year) 63.01 63.05
Spouse (share) 0.715 0.690
Working history (years) 36.25 37.19
Unemployed, share (t − 1) 0.0874 0.0402
Tertiary degree, share 0.132 0.0726
Earnings 24,705 23,702
(Continues)
565 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TABLE A3 | (Continued)
Sickness absences (t − 1)
No 1 ≤
Accrued pension (euros) 15,791 14,653
Pension wealth (euros) 328,858 304,527
Pension wealth (logs) 12.57 12.54
Δ ln(PW) 0.0667 0.0710
Δ(PW) 23,508 22,748
Psychotropic medication (t − 1) 0.0752 0.150
Medication purchases (t − 1) 6.127 10.06
Any treatment (t − 1) 0.0570 0.517
Treatment, cardio (t − 1) 0.011 0.106
Treatment, muscular (t − 1) 0.00691 0.151
Note:Monetary values are in year 2000 euros. Sickness absences in days refers to the length of sickness absence spells after the waiting period. The share of unemployed
in the previous year is identified by using information about the main type of activity of an individual within a year.
TABLE A4 | Descriptive statistics on retirement by gender and education.
Men Women Low Middle High
Risk of DI retirement 2.3% 2.2% 2.8% 2.1% 0.9%
Risk of old‐age retirement 23.8% 19.0% 24.2% 23.9% 24.6%
Share of DI retirement 57.0% 43.0% 57.0% 38.0% 5.0%
Share of old‐age retirement 55.6% 44.4% 45.5% 41.2% 13.3%
Share of population 56.3% 43.7% 45.4% 41.6% 13.1%
Note: Risk of DI retirement and risk of old‐age retirement indicate retirement risks based on different characteristic. The share of DI retirement, the share of
old‐age retirement, and the share of the population illustrate the distribution of various characteristics among retirees and the population. Before the reform
individuals aged 62 to 65 are included. After the reform individuals aged 62 to 63 are included. Low, middle and high refer to educational level. All working
sectors are included.
TABLE A5 | Descriptive statistics on retirement by year.
Year 2001 2002 2003 2004 2005 2006 2007 2008 2009
Old‐age retirement 22.1% 20.6% 15.1% 14.8% 31.0% 28.2% 25.3% 27.9% 31.1%
Disability retirement 3.3% 3.9% 3.9% 2.6% 2.0% 1.7% 1.4% 1.4% 1.3%
Note: Before the reform individuals aged 62 to 65 are included. After the reform individuals aged 62 to 63 are included. All working sectors are included.
566 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TA
BL
E
A
6
|
O
LS
es
tim
at
es
by
po
pu
la
tio
n
gr
ou
ps
,e
xc
lu
di
ng
in
di
vi
du
al
s
en
tit
le
d
to
na
tio
na
lp
en
sio
ns
.
Tr
ea
tm
en
t,
ca
rd
io
(t
−
1)
Tr
ea
tm
en
t,
m
us
cu
la
r
(t
−
1)
A
ny
tr
ea
tm
en
t(
t−
1)
M
ed
ic
at
io
n
(t
−
1)
Ps
yc
ho
tr
op
ic
m
ed
ic
at
io
n
(t
−
1)
Si
ck
ne
ss
ab
se
nc
es
(t
−
1)
Ill
he
al
th
(t
−
1)
Ye
s
N
o
Ye
s
N
o
Ye
s
N
o
8≤
7
or
le
ss
Ye
s
N
o
1≤
N
o
Ye
s
N
o
D
ep
en
de
nt
va
ria
bl
e
is
re
tir
em
en
td
ec
isi
on
Δ
ln
(P
W
)
−4
.9
28
**
*
−2
.6
37
**
*
−2
.8
79
*
−2
.6
88
**
*
−2
.5
43
**
*
−2
.6
88
**
*
−2
.7
26
**
*
−2
.6
36
**
*
−2
.9
09
**
*
−2
.6
61
**
*
−1
.1
87
−2
.8
03
**
*
−2
.8
58
**
*
−2
.5
31
**
*
(1
.2
93
)
(0
.1
77
)
(1
.4
31
)
(0
.1
77
)
(0
.5
61
)
(0
.1
84
)
(0
.2
87
)
(0
.2
23
)
(0
.5
63
)
(0
.1
86
)
(0
.6
32
)
(0
.1
83
)
(0
.2
69
)
(0
.2
32
)
ln
(P
W
)
−0
.2
61
*
0.
06
69
**
*
0.
13
0
0.
06
12
**
*
0.
03
36
0.
06
47
**
*
0.
05
02
0.
06
56
**
*
0.
04
24
0.
06
26
**
*
0.
05
83
0.
06
09
**
*
0.
04
38
0.
07
06
**
*
(0
.1
17
)
(0
.0
16
0)
(0
.1
34
)
(0
.0
15
9)
(0
.0
48
0)
(0
.0
16
7)
(0
.0
26
9)
(0
.0
19
6)
(0
.0
58
3)
(0
.0
16
5)
(0
.0
50
7)
(0
.0
16
7)
(0
.0
25
7)
(0
.0
20
1)
FR
A
0.
24
1
0.
25
4*
**
0.
17
2
0.
25
2*
**
0.
21
2*
**
0.
25
5*
**
0.
22
5*
**
0.
25
9*
**
0.
11
9
0.
26
0*
**
0.
18
5*
**
0.
26
1*
**
0.
22
7*
**
0.
26
3*
**
(0
.1
39
)
(0
.0
16
4)
(0
.1
53
)
(0
.0
16
4)
(0
.0
57
0)
(0
.0
17
0)
(0
.0
31
1)
(0
.0
19
7)
(0
.0
72
0)
(0
.0
16
8)
(0
.0
50
2)
(0
.0
17
3)
(0
.0
29
2)
(0
.0
20
1)
p‐
va
lu
e
Δ
ln
(P
W
)
0.
03
9
0.
87
7
0.
80
1
0.
80
2
0.
66
3
0.
01
2
0.
35
3
p‐
va
lu
e
ln
(P
W
)
0.
00
1
0.
55
3
0.
52
9
0.
64
2
0.
72
8
0.
95
9
0.
40
7
p‐
va
lu
e
FR
A
0.
91
2
0.
54
5
0.
45
9
0.
34
6
0.
04
6
0.
13
7
0.
30
5
O
bs
er
va
tio
ns
19
4
81
82
19
3
81
83
91
8
74
58
28
59
55
17
64
2
77
34
97
6
74
00
31
58
52
18
R‐
sq
ua
re
d
0.
48
4
0.
32
5
0.
51
1
0.
32
5
0.
32
3
0.
32
9
0.
31
7
0.
33
3
0.
34
8
0.
32
8
0.
35
9
0.
32
5
0.
31
8
0.
33
5
FE
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
In
di
vi
du
al
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
Sp
ou
se
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
H
ea
lth
co
nt
ro
ls
t−
1
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
ot
e:
Th
e
nu
ll
hy
po
th
es
is
fo
rt
he
p‐
va
lu
es
is
th
at
th
e
di
ffe
re
nc
e
be
tw
ee
n
co
ef
fic
ie
nt
se
qu
al
sz
er
o.
Δ
ln
(P
W
)
is
th
e
fin
an
ci
al
in
ce
nt
iv
es
to
po
stp
on
e
re
tir
em
en
t(
i.e
.t
he
re
la
tiv
e
ch
an
ge
in
pe
ns
io
n
w
ea
lth
,P
W
,w
he
n
re
tir
em
en
ti
s
po
stp
on
ed
by
1
ye
ar
).
Si
ck
ne
ss
ab
se
nc
es
in
di
ca
te
sd
ay
sa
fte
rt
he
w
ai
tin
g
pe
rio
d.
Th
os
ee
nt
itl
ed
to
na
tio
na
lp
en
sio
ns
ar
ee
xc
lu
de
d
fro
m
th
ea
na
ly
sis
.R
ob
us
ts
ta
nd
ar
d
er
ro
rs
in
pa
re
nt
he
se
s*
**
p
<
0.
00
1,
**
p
<
0.
01
,*
p
<
0.
05
.
567 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TA
BL
E
A
7
|
O
LS
es
tim
at
es
by
po
pu
la
tio
n
gr
ou
ps
,e
xc
lu
di
ng
un
em
pl
oy
ed
.
Tr
ea
tm
en
t,
ca
rd
io
(t
−
1)
Tr
ea
tm
en
t,
m
us
cu
la
r
(t
−
1)
A
ny
tr
ea
tm
en
t(
t−
1)
M
ed
ic
at
io
n
(t
−
1)
Ps
yc
ho
tr
op
ic
m
ed
ic
at
io
n
(t
−
1)
Si
ck
ne
ss
ab
se
nc
es
(t
−
1)
Ill
he
al
th
(t
−
1)
Ye
s
N
o
Ye
s
N
o
Ye
s
N
o
8≤
7
or
le
ss
Ye
s
N
o
1≤
N
o
Ye
s
N
o
D
ep
en
de
nt
va
ria
bl
e
is
re
tir
em
en
td
ec
isi
on
Δ
ln
(P
W
)
−2
.4
48
*
−1
.2
88
**
*
−3
.0
37
**
−1
.2
85
**
*
−1
.8
76
**
*
−1
.2
50
**
*
−1
.1
95
**
*
−1
.3
83
**
*
−1
.4
77
**
−1
.2
97
**
*
−0
.8
89
*
−1
.3
75
**
*
−1
.3
94
**
*
−1
.2
71
**
*
(1
.0
74
)
(0
.1
38
)
(0
.9
81
)
(0
.1
38
)
(0
.4
01
)
(0
.1
46
)
(0
.2
25
)
(0
.1
73
)
(0
.4
57
)
(0
.1
44
)
(0
.4
47
)
(0
.1
44
)
(0
.2
12
)
(0
.1
80
)
ln
(P
W
)
−0
.1
58
0.
04
46
**
*
0.
04
41
0.
04
23
**
*
0.
01
26
0.
04
64
**
*
0.
04
89
*
0.
03
62
**
0.
03
82
0.
04
28
**
*
0.
04
24
0.
04
27
**
*
0.
04
53
*
0.
03
78
**
(0
.0
83
6)
(0
.0
11
4)
(0
.0
98
2)
(0
.0
11
3)
(0
.0
36
2)
(0
.0
11
8)
(0
.0
19
0)
(0
.0
14
0)
(0
.0
41
4)
(0
.0
11
7)
(0
.0
36
0)
(0
.0
11
8)
(0
.0
18
2)
(0
.0
14
4)
FR
A
0.
27
3*
0.
24
0*
**
0.
06
12
0.
24
4*
**
0.
17
9*
**
0.
24
6*
**
0.
25
6*
**
0.
23
0*
**
0.
22
1*
**
0.
24
0*
**
0.
15
1*
**
0.
25
2*
**
0.
25
4*
**
0.
23
1*
**
(0
.1
12
)
(0
.0
13
5)
(0
.1
16
)
(0
.0
13
4)
(0
.0
45
4)
(0
.0
14
0)
(0
.0
22
8)
(0
.0
16
6)
(0
.0
51
2)
(0
.0
13
9)
(0
.0
39
1)
(0
.0
14
2)
(0
.0
21
6)
(0
.0
17
0)
p‐
va
lu
e
Δ
ln
(P
W
)
0.
22
7
0.
05
1
0.
13
5
0.
50
5
0.
69
9
0.
29
2
0.
65
6
p‐
va
lu
e
ln
(P
W
)
0.
00
7
0.
98
3
0.
36
6
0.
58
9
0.
91
3
0.
99
4
0.
74
5
p‐
va
lu
e
FR
A
0.
73
6
0.
08
5
0.
14
5
0.
35
7
0.
71
8
0.
01
3
0.
39
7
O
bs
er
va
tio
ns
25
8
11
,6
04
29
9
11
,5
63
13
31
10
,5
31
41
24
77
38
96
7
10
,8
95
14
81
10
,3
81
45
50
73
12
R‐
sq
ua
re
d
0.
45
7
0.
31
4
0.
40
7
0.
31
5
0.
30
2
0.
31
8
0.
31
1
0.
32
0
0.
30
4
0.
31
9
0.
34
0
0.
31
4
0.
31
0
0.
32
1
FE
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
In
di
vi
du
al
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
Sp
ou
se
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
H
ea
lth
co
nt
ro
ls
t−
1
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
ot
e:
Th
e
nu
ll
hy
po
th
es
is
fo
rt
he
p‐
va
lu
es
is
th
at
th
e
di
ffe
re
nc
e
be
tw
ee
n
co
ef
fic
ie
nt
se
qu
al
sz
er
o.
Δ
ln
(P
W
)
is
th
e
fin
an
ci
al
in
ce
nt
iv
es
to
po
stp
on
e
re
tir
em
en
t(
i.e
.t
he
re
la
tiv
e
ch
an
ge
in
pe
ns
io
n
w
ea
lth
,P
W
,w
he
n
re
tir
em
en
ti
s
po
stp
on
ed
by
1
ye
ar
).
Si
ck
ne
ss
ab
se
nc
es
in
di
ca
te
s
da
ys
af
te
r
th
e
w
ai
tin
g
pe
rio
d.
In
di
vi
du
al
s
w
ho
w
er
e
m
ai
nl
y
un
em
pl
oy
ed
in
th
e
pr
ev
io
us
ye
ar
ar
e
ex
cl
ud
ed
fro
m
th
e
an
al
ys
is.
Ro
bu
st
sta
nd
ar
d
er
ro
rs
in
pa
re
nt
he
se
s
**
*p
<
0.
00
1,
**
p
<
0.
01
,*
p
<
0.
05
.
568 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TA
BL
E
A
8
|
O
LS
es
tim
at
es
by
po
pu
la
tio
n
gr
ou
ps
,1
ye
ar
la
gg
ed
in
co
m
e.
Tr
ea
tm
en
t,
ca
rd
io
(t
−
1)
Tr
ea
tm
en
t,
m
us
cu
la
r
(t
−
1)
A
ny
tr
ea
tm
en
t(
t−
1)
M
ed
ic
at
io
n
(t
−
1)
Ps
yc
ho
tr
op
ic
m
ed
ic
at
io
n
(t
−
1)
Si
ck
ne
ss
ab
se
nc
es
(t
−
1)
Ill
he
al
th
(t
−
1)
Ye
s
N
o
Ye
s
N
o
Ye
s
N
o
8≤
7
or
le
ss
Ye
s
N
o
1≤
N
o
Ye
s
N
o
D
ep
en
de
nt
va
ria
bl
e
is
re
tir
em
en
td
ec
isi
on
Δ
ln
(P
W
)
−3
.5
48
**
*
−2
.2
67
**
*
−2
.8
48
**
*
−2
.2
83
**
*
−2
.3
35
**
*
−2
.2
96
**
*
−2
.2
49
**
*
−2
.3
08
**
*
−2
.3
58
**
*
−2
.2
85
**
*
−1
.1
95
**
−2
.4
11
**
*
−2
.3
93
**
*
−2
.2
28
**
*
(1
.0
06
)
(0
.1
26
)
(0
.8
47
)
(0
.1
27
)
(0
.3
57
)
(0
.1
34
)
(0
.2
07
)
(0
.1
58
)
(0
.4
00
)
(0
.1
32
)
(0
.4
22
)
(0
.1
31
)
(0
.1
93
)
(0
.1
65
)
ln
(P
W
)
−0
.1
30
0.
05
26
**
*
0.
05
56
0.
05
05
**
*
0.
02
54
0.
05
38
**
*
0.
05
51
**
0.
04
45
**
*
0.
06
31
0.
04
91
**
*
0.
04
40
0.
05
13
**
*
0.
05
40
**
0.
04
49
**
(0
.0
79
8)
(0
.0
10
9)
(0
.0
92
9)
(0
.0
10
8)
(0
.0
33
9)
(0
.0
11
4)
(0
.0
18
2)
(0
.0
13
4)
(0
.0
37
6)
(0
.0
11
3)
(0
.0
34
8)
(0
.0
11
3)
(0
.0
17
4)
(0
.0
13
8)
FR
A
0.
24
7*
0.
23
7*
**
0.
04
82
0.
24
0*
**
0.
18
3*
**
0.
24
2*
**
0.
25
3*
**
0.
22
7*
**
0.
21
7*
**
0.
23
7*
**
0.
15
5*
**
0.
24
7*
**
0.
25
2*
**
0.
22
7*
**
(0
.1
00
)
(0
.0
12
8)
(0
.1
13
)
(0
.0
12
8)
(0
.0
42
2)
(0
.0
13
4)
(0
.0
21
9)
(0
.0
15
7)
(0
.0
47
6)
(0
.0
13
3)
(0
.0
37
8)
(0
.0
13
5)
(0
.0
20
6)
(0
.0
16
2)
p‐
va
lu
e
Δ
ln
(P
W
)
0.
16
0.
46
8
0.
91
7
0.
81
9
0.
86
0.
00
5
0.
51
3
p‐
va
lu
e
ln
(P
W
)
0.
01
2
0.
95
2
0.
41
8
0.
63
7
0.
71
5
0.
83
9
0.
68
1
p‐
va
lu
e
FR
A
0.
91
2
0.
06
4
0.
17
7
0.
32
2
0.
68
9
0.
02
0.
32
7
O
bs
er
va
tio
ns
28
7
12
,6
36
31
3
12
,6
10
14
47
11
,4
76
45
01
84
22
10
88
11
,8
35
15
43
11
,3
80
49
90
79
33
R‐
sq
ua
re
d
0.
44
5
0.
32
1
0.
41
3
0.
32
2
0.
32
0
0.
32
3
0.
31
4
0.
32
8
0.
31
3
0.
32
6
0.
34
4
0.
32
2
0.
31
5
0.
32
8
FE
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
In
di
vi
du
al
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
Sp
ou
se
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
H
ea
lth
co
nt
ro
ls
t−
1
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
ot
e:
Th
e
nu
ll
hy
po
th
es
is
fo
rt
he
p‐
va
lu
es
is
th
at
th
e
di
ffe
re
nc
e
be
tw
ee
n
th
e
co
ef
fic
ie
nt
se
qu
al
sz
er
o.
Δ
ln
(P
W
)
is
th
e
re
la
tiv
e
ch
an
ge
in
pe
ns
io
n
w
ea
lth
w
he
n
re
tir
em
en
ti
sp
os
tp
on
ed
by
1
ye
ar
.P
en
sio
n
w
ea
lth
is
ca
lc
ul
at
ed
us
in
g
1
ye
ar
la
gg
ed
in
co
m
es
.S
ic
kn
es
s
ab
se
nc
es
in
di
ca
te
s
da
ys
af
te
r
th
e
w
ai
tin
g
pe
rio
d.
Ro
bu
st
sta
nd
ar
d
er
ro
rs
in
pa
re
nt
he
se
s
**
*p
<
0.
00
1,
**
p
<
0.
01
,*
p
<
0.
05
.
569 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TA
BL
E
A
9
|
O
LS
es
tim
at
es
by
po
pu
la
tio
n
gr
ou
ps
in
cl
ud
in
g
di
sa
bi
lit
y
re
tir
em
en
ts.
Tr
ea
tm
en
t,
ca
rd
io
(t
−
1)
Tr
ea
tm
en
t,
m
us
cu
la
r
(t
−
1)
A
ny
tr
ea
tm
en
t(
t−
1)
M
ed
ic
at
io
n
(t
−
1)
Ps
yc
ho
tr
op
ic
m
ed
ic
at
io
n
(t
−
1)
Si
ck
ne
ss
ab
se
nc
es
(t
−
1)
Ill
he
al
th
(t
−
1)
Ye
s
N
o
Ye
s
N
o
Ye
s
N
o
8≤
7
or
le
ss
Ye
s
N
o
1≤
N
o
Ye
s
N
o
D
ep
en
de
nt
va
ria
bl
e
is
re
tir
em
en
td
ec
isi
on
Δ
ln
(P
W
)
−3
.1
03
**
−2
.1
25
**
*
−3
.4
77
**
*
−2
.1
29
**
*
−2
.4
68
**
*
−2
.1
18
**
*
−2
.0
43
**
*
−2
.2
14
**
*
−1
.9
97
**
*
−2
.1
65
**
*
−1
.8
27
**
*
−2
.2
73
**
*
−2
.2
00
**
*
−2
.1
03
**
*
(1
.0
11
)
(0
.1
30
)
(0
.8
56
)
(0
.1
30
)
(0
.3
79
)
(0
.1
37
)
(0
.2
13
)
(0
.1
61
)
(0
.4
18
)
(0
.1
35
)
(0
.4
54
)
(0
.1
34
)
(0
.2
00
)
(0
.1
68
)
ln
(P
W
)
−0
.1
65
0.
04
72
**
*
0.
03
97
0.
04
51
**
*
−0
.0
02
78
0.
04
98
**
*
0.
04
19
*
0.
04
36
**
0.
05
06
0.
04
41
**
*
−0
.0
04
04
0.
05
06
**
*
0.
04
04
*
0.
04
45
**
(0
.0
84
4)
(0
.0
11
2)
(0
.0
97
6)
(0
.0
11
1)
(0
.0
35
3)
(0
.0
11
6)
(0
.0
18
7)
(0
.0
13
8)
(0
.0
39
1)
(0
.0
11
6)
(0
.0
38
0)
(0
.0
11
5)
(0
.0
18
0)
(0
.0
14
1)
FR
A
0.
20
0*
0.
24
0*
**
0.
01
42
0.
24
2*
**
0.
16
6*
**
0.
24
6*
**
0.
24
3*
**
0.
23
2*
**
0.
21
2*
**
0.
23
8*
**
0.
14
0*
**
0.
25
4*
**
0.
24
5*
**
0.
23
1*
**
(0
.0
98
2)
(0
.0
13
1)
(0
.1
12
)
(0
.0
13
0)
(0
.0
43
3)
(0
.0
13
6)
(0
.0
22
6)
(0
.0
15
9)
(0
.0
48
6)
(0
.0
13
5)
(0
.0
38
9)
(0
.0
13
7)
(0
.0
21
3)
(0
.0
16
4)
p‐
va
lu
e
Δ
ln
(P
W
)
0.
29
2
0.
09
0.
37
8
0.
52
0.
69
6
0.
33
9
0.
70
9
p‐
va
lu
e
ln
(P
W
)
0.
00
6
0.
95
2
0.
15
0.
94
1
0.
87
1
0.
16
3
0.
85
5
p‐
va
lu
e
FR
A
0.
65
8
0.
02
9
0.
07
4
0.
69
7
0.
59
3
0.
00
5
0.
59
7
O
bs
er
va
tio
ns
32
5
12
,9
19
34
8
12
,8
96
15
66
11
,6
78
46
91
85
53
11
66
12
,0
78
17
91
11
,4
53
52
45
79
99
R‐
sq
ua
re
d
0.
33
9
0.
27
5
0.
30
4
0.
27
6
0.
22
5
0.
28
5
0.
24
5
0.
29
4
0.
21
5
0.
28
4
0.
16
6
0.
30
5
0.
23
6
0.
30
6
FE
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
In
di
vi
du
al
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
Sp
ou
se
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
H
ea
lth
co
nt
ro
ls
t−
1
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
ot
e:
Th
e
nu
ll
hy
po
th
es
is
fo
rt
he
p‐
va
lu
es
is
th
at
th
e
di
ffe
re
nc
e
be
tw
ee
n
co
ef
fic
ie
nt
se
qu
al
sz
er
o.
Δ
ln
(P
W
)
is
th
e
fin
an
ci
al
in
ce
nt
iv
es
to
po
stp
on
e
re
tir
em
en
t(
i.e
.,
th
e
re
la
tiv
e
ch
an
ge
in
pe
ns
io
n
w
ea
lth
,P
W
,w
he
n
re
tir
em
en
ti
s
po
stp
on
ed
by
1
ye
ar
).
Si
ck
ne
ss
ab
se
nc
es
in
di
ca
te
sd
ay
sa
fte
rt
he
w
ai
tin
g
pe
rio
d.
Pe
ns
io
n
w
ea
lth
an
d
th
e
in
ce
nt
iv
es
ar
e
ca
lc
ul
at
ed
ac
co
rd
in
g
to
Eq
ua
tio
n
(4
).
Ro
bu
st
st
an
da
rd
er
ro
rs
in
pa
re
nt
he
se
s*
**
p
<
0.
00
1,
**
p
<
0.
01
,
*p
<
0.
05
.
570 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TA
BL
E
A
10
|
2S
LS
es
tim
at
es
by
po
pu
la
tio
n
gr
ou
ps
(m
ai
n
sa
m
pl
e)
,1
ye
ar
la
gg
ed
in
co
m
e.
Tr
ea
tm
en
t,
ca
rd
io
(t
−
1)
Tr
ea
tm
en
t,
m
us
cu
la
r
(t
−
1)
A
ny
tr
ea
tm
en
t(
t−
1)
M
ed
ic
at
io
n
(t
−
1)
Ps
yc
ho
tr
op
ic
m
ed
ic
at
io
n
(t
−
1)
Si
ck
ne
ss
ab
se
nc
es
(t
−
1)
Ill
he
al
th
(t
−
1)
Ye
s
N
o
Ye
s
N
o
Ye
s
N
o
8≤
7
or
le
ss
Ye
s
N
o
1≤
N
o
Ye
s
N
o
D
ep
en
de
nt
va
ria
bl
e
is
re
tir
em
en
td
ec
isi
on
Δ
ln
(P
W
)
−3
.5
43
**
*
−2
.3
43
**
*
−2
.9
68
**
*
−2
.3
57
**
*
−2
.4
57
**
*
−2
.3
65
**
*
−2
.3
14
**
*
−2
.3
88
**
*
−2
.3
95
**
*
−2
.3
64
**
*
−1
.2
31
**
−2
.4
91
**
*
−2
.4
60
**
*
−2
.3
07
**
*
(0
.9
10
)
(0
.1
30
)
(0
.8
01
)
(0
.1
30
)
(0
.3
67
)
(0
.1
37
)
(0
.2
10
)
(0
.1
63
)
(0
.3
94
)
(0
.1
36
)
(0
.4
28
)
(0
.1
35
)
(0
.1
97
)
(0
.1
70
)
ln
(P
W
)
−0
.1
28
0.
05
13
**
*
0.
05
14
0.
04
92
**
*
0.
02
58
0.
05
23
**
*
0.
05
31
**
0.
04
35
**
0.
06
15
0.
04
78
**
*
0.
04
29
0.
04
99
**
*
0.
05
20
**
0.
04
40
**
(0
.0
72
3)
(0
.0
11
0)
(0
.0
84
9)
(0
.0
10
9)
(0
.0
33
5)
(0
.0
11
4)
(0
.0
18
2)
(0
.0
13
5)
(0
.0
36
8)
(0
.0
11
3)
(0
.0
34
3)
(0
.0
11
4)
(0
.0
17
4)
(0
.0
13
8)
FR
A
0.
24
6*
*
0.
23
6*
**
0.
04
38
0.
24
0*
**
0.
18
5*
**
0.
24
1*
**
0.
25
3*
**
0.
22
6*
**
0.
21
8*
**
0.
23
6*
**
0.
15
5*
**
0.
24
7*
**
0.
25
2*
**
0.
22
6*
**
(0
.0
90
7)
(0
.0
12
8)
(0
.1
03
)
(0
.0
12
8)
(0
.0
41
5)
(0
.0
13
3)
(0
.0
21
7)
(0
.0
15
7)
(0
.0
46
4)
(0
.0
13
2)
(0
.0
37
1)
(0
.0
13
5)
(0
.0
20
5)
(0
.0
16
2)
p‐
va
lu
e
Δ
ln
(P
W
)
0.
19
2
0.
45
1
0.
81
5
0.
78
3
0.
93
9
0.
00
5
0.
55
7
p‐
va
lu
e
ln
(P
W
)
0.
01
5
0.
97
9
0.
45
5
0.
67
2
0.
72
1
0.
84
5
0.
72
p‐
va
lu
e
FR
A
0.
91
2
0.
06
0.
19
5
0.
30
9
0.
71
4
0.
02
1
0.
31
6
O
bs
er
va
tio
ns
28
7
12
,6
36
31
3
12
,6
10
14
47
11
,4
76
45
01
84
22
10
88
11
,8
35
15
43
11
,3
80
49
90
79
33
R‐
sq
ua
re
d
0.
43
8
0.
31
9
0.
41
2
0.
32
0
0.
32
0
0.
32
2
0.
31
1
0.
32
7
0.
31
2
0.
32
4
0.
34
3
0.
32
0
0.
31
3
0.
32
7
Kl
ei
be
rg
en
‐P
aa
p
10
,4
40
57
,4
95
34
90
57
,7
72
17
81
90
,0
77
15
,5
47
52
,9
54
19
,7
60
50
,2
34
86
39
52
,0
18
18
,5
90
47
,4
10
F
sta
tis
tic
s
Fi
rs
ts
ta
ge
:D
ep
en
de
nt
va
ria
bl
e
is
re
la
tiv
e
ch
an
ge
in
pe
ns
io
n
w
ea
lth
ca
lc
ul
at
ed
us
in
g
1
ye
ar
la
gg
ed
in
co
m
e
ln
P
W
t+
1
y t
,…
))
−
ln
P
W
t
y t−
1,
…
))
)
Fi
rs
ts
ta
ge
0.
98
0.
96
3
0.
97
2
0.
96
4
0.
95
2
0.
96
5
0.
96
5
0.
96
2
0.
98
0.
96
2
0.
96
8
0.
96
3
0.
96
7
0.
96
1
r‐s
qu
ar
ed
FE
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
In
di
vi
du
al
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
Sp
ou
se
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
H
ea
lth
co
nt
ro
ls
t−
1
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
ot
e:
Th
e
nu
ll
hy
po
th
es
is
fo
rt
he
p‐
va
lu
es
is
th
at
th
e
di
ffe
re
nc
e
be
tw
ee
n
co
ef
fic
ie
nt
se
qu
al
sz
er
o.
Δ
ln
(P
W
)
is
th
e
fin
an
ci
al
in
ce
nt
iv
es
to
po
stp
on
e
re
tir
em
en
t(
i.e
.t
he
re
la
tiv
e
ch
an
ge
in
pe
ns
io
n
w
ea
lth
,P
W
,w
he
n
re
tir
em
en
ti
s
po
stp
on
ed
by
1
ye
ar
).
Si
ck
ne
ss
ab
se
nc
es
in
di
ca
te
s
da
ys
af
te
r
th
e
w
ai
tin
g
pe
rio
d.
Ro
bu
st
sta
nd
ar
d
er
ro
rs
in
pa
re
nt
he
se
s
**
*p
<
0.
00
1,
**
p
<
0.
01
,*
p
<
0.
05
.
571 of 598
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
TA
BL
E
A
11
|
2S
LS
es
tim
at
es
by
po
pu
la
tio
n
gr
ou
ps
,2
ye
ar
la
gg
ed
in
co
m
e.
Tr
ea
tm
en
t,
ca
rd
io
(t
−
1)
Tr
ea
tm
en
t,
m
us
cu
la
r
(t
−
1)
A
ny
tr
ea
tm
en
t(
t−
1)
M
ed
ic
at
io
n
(t
−
1)
Ps
yc
ho
tr
op
ic
m
ed
ic
at
io
n
(t
−
1)
Si
ck
ne
ss
ab
se
nc
es
(t
−
1)
Ill
he
al
th
(t
−
1)
Ye
s
N
o
Ye
s
N
o
Ye
s
N
o
8≤
7
or
le
ss
Ye
s
N
o
1≤
N
o
Ye
s
N
o
D
ep
en
de
nt
va
ria
bl
e
is
re
tir
em
en
td
ec
isi
on
Δ
ln
(P
W
)
−3
.5
09
**
*
−2
.8
22
**
*
−2
.8
90
**
−2
.8
38
**
*
−3
.0
89
**
*
−2
.8
25
**
*
−2
.7
85
**
*
−2
.8
61
**
*
−2
.8
21
**
*
−2
.8
41
**
*
−0
.9
54
−3
.0
29
**
*
−2
.9
21
**
*
−2
.7
94
**
*
(1
.0
61
)
(0
.1
48
)
(0
.9
56
)
(0
.1
48
)
(0
.4
43
)
(0
.1
55
)
(0
.2
37
)
(0
.1
87
)
(0
.4
48
)
(0
.1
55
)
(0
.5
23
)
(0
.1
53
)
(0
.2
23
)
(0
.1
94
)
ln
(P
W
)
−0
.1
39
0.
04
82
**
*
0.
07
15
0.
04
56
**
*
0.
03
54
0.
04
80
**
*
0.
04
86
*
0.
04
14
**
0.
04
03
0.
04
59
**
*
0.
04
94
0.
04
60
**
*
0.
04
71
**
0.
04
29
**
(0
.0
79
3)
(0
.0
11
6)
(0
.0
90
8)
(0
.0
11
5)
(0
.0
35
0)
(0
.0
12
1)
(0
.0
19
0)
(0
.0
14
3)
(0
.0
38
3)
(0
.0
12
0)
(0
.0
36
0)
(0
.0
12
0)
(0
.0
18
1)
(0
.0
14
8)
FR
A
0.
26
5*
*
0.
23
8*
**
−0
.0
43
7
0.
24
3*
**
0.
19
8*
**
0.
24
2*
**
0.
25
6*
**
0.
22
7*
**
0.
26
4*
**
0.
23
5*
**
0.
14
9*
**
0.
25
0*
**
0.
25
7*
**
0.
22
6*
**
(0
.0
96
3)
(0
.0
13
9)
(0
.1
28
)
(0
.0
13
8)
(0
.0
44
6)
(0
.0
14
5)
(0
.0
23
8)
(0
.0
17
0)
(0
.0
49
0)
(0
.0
14
4)
(0
.0
41
2)
(0
.0
14
6)
(0
.0
22
5)
(0
.0
17
5)
p‐
va
lu
e
Δ
ln
(P
W
)
0.
52
2
0.
95
7
0.
57
4
0.
80
1
0.
96
6
<
0.
00
1
0.
66
9
p‐
va
lu
e
ln
(P
W
)
0.
02
0.
77
7
0.
73
3
0.
76
0.
88
8
0.
92
7
0.
85
6
p‐
va
lu
e
FR
A
0.
78
3
0.
02
5
0.
34
6
0.
31
2
0.
57
1
0.
02
1
0.
28
6
O
bs
er
va
tio
ns
26
4
11
,4
48
28
2
11
,4
30
13
36
10
,3
76
41
57
75
55
99
7
10
,7
15
14
01
10
,3
11
45
99
71
13
R‐
sq
ua
re
d
0.
44
6
0.
30
4
0.
41
0
0.
30
5
0.
31
0
0.
30
6
0.
30
0
0.
31
0
0.
32
0
0.
30
7
0.
33
0
0.
30
5
0.
30
2
0.
31
0
Kl
ei
be
rg
en
‐P
aa
p
23
4
12
,1
17
52
8
12
,0
58
80
1
12
,8
07
64
96
67
07
59
4
12
,8
84
85
9
11
,3
64
45
49
80
47
F
sta
tis
tic
s
Fi
rs
ts
ta
ge
:D
ep
en
de
nt
va
ria
bl
e
is
re
la
tiv
e
ch
an
ge
in
pe
ns
io
n
w
ea
lth
ca
lc
ul
at
ed
us
in
g
2
ye
ar
la
gg
ed
in
co
m
e
ln
P
W
t+
1
y t−
1,
…
))
−
ln
P
W
t
y t−
2,
…
))
)
Fi
rs
ts
ta
ge
0.
80
8
0.
82
9
0.
83
5
0.
82
9
0.
75
4
0.
83
9
0.
83
6
0.
82
4
0.
84
4
0.
82
6
0.
79
0.
83
2
0.
82
6
0.
83
r‐s
qu
ar
ed
FE
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
In
di
vi
du
al
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
Sp
ou
se
co
nt
ro
ls
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
H
ea
lth
co
nt
ro
ls
t−
1
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
O
N
ot
e:
Th
e
nu
ll
hy
po
th
es
is
fo
rt
he
p‐
va
lu
es
is
th
at
th
e
di
ffe
re
nc
e
be
tw
ee
n
co
ef
fic
ie
nt
se
qu
al
sz
er
o.
Δ
ln
(P
W
)
is
th
e
fin
an
ci
al
in
ce
nt
iv
es
to
po
stp
on
e
re
tir
em
en
t(
i.e
.t
he
re
la
tiv
e
ch
an
ge
in
pe
ns
io
n
w
ea
lth
,P
W
,w
he
n
re
tir
em
en
ti
s
po
stp
on
ed
by
1
ye
ar
).
Si
ck
ne
ss
ab
se
nc
es
in
di
ca
te
s
da
ys
af
te
r
th
e
w
ai
tin
g
pe
rio
d.
Ro
bu
st
sta
nd
ar
d
er
ro
rs
in
pa
re
nt
he
se
s
**
*p
<
0.
00
1,
**
p
<
0.
01
,*
p
<
0.
05
.
572 of 598 Health Economics, 2025
 10991050, 2025, 3, D
ow
nloaded from
 https://onlinelibrary.w
iley.com
/doi/10.1002/hec.4917 by U
niversity of Turku, W
iley O
nline Library on [12/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on W
iley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License