ArticlePDF Available

Biased Health Perceptions and Risky Health Behaviors-Theory and Evidence

Authors:

Abstract and Figures

This paper investigates the role of biased health perceptions as a potential driving force of risky health behaviors. We define absolute and relative health perception biases, illustrate their measurement in surveys and provide evidence on their relevance. Next, we decompose the theoretical effect into its extensive and intensive margin: When the extensive margin dominates, people (wrongly) believe they are healthy enough to "afford" unhealthy behavior. Finally, using three population surveys, we provide robust empirical evidence that respondents who overestimate their health are less likely to exercise and sleep enough, but more likely to eat unhealthily and drink alcohol daily.
Content may be subject to copyright.
Biased Health Perceptions and Risky Health
Behaviors—Theory and Evidence
Patrick Arni
University of Bristol
Davide Dragone
University of Bologna
Lorenz Goette
University of Bonn
Nicolas R. Ziebarth
Cornell University *
November 20, 2020
Abstract
This paper investigates the role of biased health perceptions as a potential driving force of
risky health behaviors. We define absolute and relative health perception biases, illustrate
their measurement in surveys and provide evidence on their relevance. Next, we decom-
pose the theoretical effect into its extensive and intensive margin: When the extensive mar-
gin dominates, people (wrongly) believe they are healthy enough to “afford” unhealthy
behavior. Finally, using three population surveys, we provide robust empirical evidence
that respondents who overestimate their health are less likely to exercise and sleep enough,
but more likely to eat unhealthily and drink alcohol daily.
Keywords: health bias, health perceptions, subjective beliefs, overconfidence, overopti-
mism, risky behavior, smoking, obesity, exercising, SF12, SAH, BASE-II
JEL classification: D11, D83, D91, I12, P46
We would like to thank Teresa Bago d’Uva, Emily Beam, Kitt Carpenter, John Cawley, Davide Cesarini, Resul
Cesur, Owen O’Donnell, Fabrice Etil´
e, Osea Giuntella, Glenn Harrison, Ben Hansen, Hendrik J ¨
uerges, Jonathan
Ketcham, Nadine Ketel, Gaurav Khanna, Audrey Laporte, Fabian Lange, Nathalie Mathieu-Bolh, Taryn Morris-
sey, Robert Nuscheler, Reto Odermatt, Ricardo Perez-Truglia, Gregor Pfeifer, Pia Pinger, Aldo Rustichini, Joe Sabia,
Luis Santos Pinto, Tom Siedler, Rodrigo Soares, Sara Solnick, Pascal St-Amour, Alois Stutzer, Justin Sydnor, Harald
Tauchmann, Erdal Tekin, Christian Traxler, Gerard van den Berg, Ben Vollaard, Justin White, V´
era Zabrodina. In
particular we thank our discussants Matt Harris and Nathan Kettlewell for excellent comments and suggestions.
Moreover, we thank conference participants at the the 2019 iHEA World Congress in Basel, the 2019 Workshop
on the Economics of Risky Behavior in Bologna, the 2018 ASHEcon meetings in Atlanta, 2017 Bristol Workshop
on Economic Policy Interventions and Behaviour, the 2017 Risky Health Behaviors Workshop in Hamburg, the
2014 iHEA/ECHE conference in Dublin as well as seminar participants at the University of Basel, University of
Reno, the Universit of Vermont, the Berlin Network of Labor Market Research (BeNA), and The Institute on Health
Economics, Health Behaviors and Disparities at Cornell University. We take responsibility for all remaining er-
rors in and shortcomings of the article. This article uses data from the Berlin Aging Study II (BASE-II) which has
been supported by the German Federal Ministry of Education and Research and Research under grant numbers
16SV5537/16SV55837/16SV5538/16SV5536K/01UW0808. We also would like to thank Peter Eibich and Katrin
Schaar for excellent support with the BASE-II data, David Richter for his excellent support with the SOEP-IP data
as well as Gert Wagner for his overall support of this research and with the BASE-II and SOEP-IP data. The research
reported in this paper is not the result of a for-pay consulting relationship. The responsibility for the contents of
this publication lies with its authors. Our employers do not have a financial interest in the topic of the paper that
might constitute a conflict of interest. The project has been exempt from Cornell IRB review under ID # 1309004122.
*Corresponding author: Cornell University, Department of Policy Analysis and Management (PAM), 426
Kennedy Hall, Ithaca, NY 14850, USA, phone: +1-(607)255-1180, e-mail: nrz2@cornell.edu
1 Introduction
“I am so healthy I can do dangerous things and still be unlined, as yet unscathed, and beautiful.”
(Nancy Etcoff, psychologist at Harvard Medical School, in: Survival of the Prettiest: The Science of
Beauty, 2011)
Can risky health behavior be “optimal”? While most non-economists would immediately
refute this thought, economists would probably answer “it depends” and observe that, al-
though drinking alcohol or eating junk food can be detrimental to health, people pursue those
activities because they are also pleasurable. For economists, the optimality of a risky health
behavior ultimately depends on the associated cost and benefits as perceived by the individual
(Grossman,1972;Becker and Murphy,1988;Dragone,2009;Cawley and Ruhm,2011).
According to the benchmark theoretical framework, economic agents have perfect informa-
tion and infinite computational capacity, which allows them to correctly assess costs and bene-
fits of each alternative and, accordingly, choose optimal risky behavior. Behavioral economists
have challenged this view by empirically and theoretically studying behavior that deviates
from the predictions obtained under perfect rationality and information (cf. Rabin,2013). In
health economics, influential studies have shown that consumers pick dominated health plans
and leave money on the table (Abaluck and Gruber,2011,2016;Ketcham et al.,2012,2016;
Bhargava et al.,2017;Kettlewell,2020). Research has also cited behavioral phenomena as ex-
planations for why people engage in “too much” risky health behavior. For example, the-
oretical papers model the role of hyperbolic discounting and time inconsistencies for smok-
ing and overeating (Gruber and K˝
oszegi,2001;Strulik and Trimborn,2018). In a field exper-
iment among college students, Avery et al. (2019) show that present-biased individuals are
more likely to take up commitment devices to sleep enough, which reduces their insufficient
sleep significantly. To explain the fact that many gym members overpay (Della Vigna and Mal-
mendier,2006), other field experiments find that nudges to exercise more regularly are mostly
effective in the short-run (Royer et al.,2015;Carrera et al.,2018,2020).
This paper is one of the first to investigate whether biased health perceptions could be a po-
tential driving force of risky health behaviors. The causes and consequences of biased beliefs
about own health are inherently difficult to study in causal effects frameworks, which is one
reason why the health economics literature on this topic is very scant. One of the notable ex-
ceptions is Harris (2017) who finds that people who overestimate their activity levels consume
1
more calories.1As a first main contribution, we formally introduce the concept of health per-
ception biases into the (health) economics literature and document the existence of biased health
perceptions in the population using three high quality datasets. Specifically, we propose two
individual measures of absolute and relative health perception biases. Absolute perception bi-
ases are biased perceptions solely of own health, whereas relative perception biases are biased
perceptions of own health relative to the health of the population and peers (c.f. Blanchflower
et al.,2009;Mathieu-Bolh and Wendner,2020). We show that, under plausible assumptions,
there exists a one-to-one positive mapping between these two health perception measures.
Although we deliberately choose the much broader definition and interpretation of “health
perception biases”, our measures could also be interpreted in the context of the concept of
“overconfidence” (Camerer and Lovallo,1999;Barber and Odean,2001;Burks et al.,2013;Spin-
newijn,2015;Bago d’Uva et al.,2020;Cowan,2018) or, outside economics, of overoptimism
biases (Weinstein,1980,1989;Sharot,2011) and self-esteem (Himmler and Koenig,2012). Over-
confidence implies that a person believes that her ability, performance or information is better
than it actually is (Benoˆ
ıt and Dubra,2011;Ortoleva and Snowberg,2015;Heidhues et al.,2019).
It is an intuitive notion that has, however, no unambiguous definition or measurement in the
economics literature. Studies have operationalized it as (i) overestimation of own performance,
(ii) overplacement of own performance relative to others, or (iii) overprecision of own informa-
tion or beliefs (Moore and Healy,2008).
Our definitions of absolute and relative health perception biases are akin to the first two
operationalizations of overconfidence, (i) overestimation and (ii) overplacement. However, it
is important to emphasize that we do not aim at explaining why people have biased health
perceptions, nor does this paper argue about the rationality of such beliefs. Our main goal is
to theoretically and empirically investigate biased health perceptions as a potential underlying
mechanism for health-related behavior and outcomes. To avoid confusion among the many no-
tions of overconfidence, in this paper we simply refer to “health perception biases.” Moreover,
we would like to clarify that this paper investigates the role of biased perceptions about own
health, not the role of biased perceptions about the consequences of risky behavior (cf. Viscusi,
1990;Lundborg,2007;Ziebarth,2018;Belot et al.,2019).
1In another study about health perception biases, Cawley and Philipson (1999) use life insurance data and the
differential between perceived and predicted mortality risk to show that observed empirical patterns are inconsis-
tent with standard insurance theory under perfect information. A possible explanation is that insurance mandates
make low risk individuals worse off if many market participants are overconfident (Sandroni and Squintani,2007).
2
As the quote in the epigraph demonstrates, but as also evident from the medical literature
and guidelines, humans make decisions about their health behavior not just based on the health
consequences of such behavior but also based on their personal (perceived) health status (Na-
tional Institutes of Health,2000;Maguire et al.,2000;Zwald et al.,2019). This is the core of this
paper: If an individual’s perception about her health status is biased, how could this health
perception bias possibly affect her health behavior, and under what conditions does it deviate
from behavior under correct beliefs?
Note that, to identify perception biases, the researcher must not only know the individual’s
perception of her health but also her true health status. To measure absolute health perception bi-
ases, we use objectively diagnosed health conditions (high cholesterol and high blood pressure)
and elicit how respondents’ perceptions about having such a condition deviate from the truth.
In a representative German survey, we find that 30% of the population have biased perceptions
about their high cholesterol levels.
To measure relative health perception biases, we asked respondents to rank their health sta-
tus relative to a reference group. Specifically, for the purpose of this paper, we included the
question in two high-quality surveys from Germany (one representative survey and one inter-
disciplinary survey). Comparing the elicited subjective ranking to the objective ranking within
the population health distribution produces our measure of relative health perception biases.
Consistent with the absolute measure, we find that about 30% of the respondents overestimate
their rank in the population health distribution by at least 30 ranks. That is, for instance, they
believe that they rank at the 70th percentile when they actually only rank at the 40th percentile of
the population health distribution. In particular, we find an excess mass of people who believe
that they rank between the 70th and the 90th percentile of the population health distribution.
To bound the impact of reporting errors in subjective health (cf. Lindeboom and van Doorslaer,
2004;Ziebarth,2010), we benchmark our relative health perception measure against the stan-
dard Self Assessed Health (SAH) measure as well as the 12-Item Short Form (SF12) health
survey measure. By construction, generic health measures like the SF12 include fewer system-
atic response biases as they are designed to produce unbiased continuous measures of physical
and mental health. As the researcher requires an unbiased benchmark health measure to em-
pirically identify perception biases, we test the robustness of our findings and the relevance of
reporting errors by using both the SAH and SF12 as benchmark measures while controlling for
a rich set of socio-demographics. While our true benchmark health measures for the absolute
3
health perception bias are objective but also narrowly focused on physical health, the SAH and
SF12 measures measure overall health and follow a broader concept. The empirical pattern and
population distributions of two absolute and two relative bias measures across three different
datasets yield important insights into the robustness and prevalence of such biases.
Next, we provide a theoretical framework that shows how biased health perceptions can
affect risky health behavior. The framework is simple and flexible enough to explain nonlinear
patterns between biased beliefs and risky behavior. Moreover, it highlights that biased health
perceptions affect behavior through an extensive and an intensive margin, and that these mar-
gins operate in opposite directions. When the extensive margin dominates, risky health behav-
ior and biased health perceptions are complements. This means that, the higher the perception
of own health, the more an individual engages in unhealthy behavior, such as consuming fast
food or not exercising. This is akin to saying: “Because I believe I am very healthy and can
afford it, I eat more fast food and exercise less.” On the contrary, when the intensive margin
of perceived health dominates, risky health behavior and health biases are substitutes, hence
a higher perceived health reduces risky behavior. This is akin to saying: “Because I perceive
large health costs of risky behavior, I will eat less fast food and exercise more.” Whether the
extensive or intensive margin dominates is an empirical question.
Hence, as a final contribution, we document robust statistical links between health percep-
tion biases and risky health behaviors across all three datasets. Specifically, individuals who
overestimate their health are significantly more likely to not exercise, to eat unhealthy, to be
overweight and to sleep fewer hours. The statistical relationships are robust to controlling for
socio-demographics, personality traits, cognitive skills, and risk aversion. They are also robust
across our two notions of health perception biases. In the context of our model, these results
are consistent with a dominant role of the extensive margin of health perception biases. Con-
versely, we do not find significant relationships for unbiased respondents and for those who
are pessimistic about their health, a finding that is consistent with the extensive and inten-
sive margin of health perception bias offsetting each other.2Notably, we find that smoking
is not correlated with biased health perceptions. This result is consistent with Darden (2017),
who finds that cardiovascular biomarker information at repeated health exams does not signif-
icantly alter smoking behavior. A possible explanation is that signals and information about
own health, be it objective as in Darden (2017) or perceived as in this paper, are not powerful
2As it is common in this literature, our results cannot exclude that biased health perceptions and risky health
behaviors are linked via unobservables, for instance genes (Linn´
er et al.,2019), or that the causality runs from risky
behaviors to perception biases in the form of self-serving biases (B´
enabou and Tirole,2002).
4
drivers of smoking behavior, possibly because its addictive nature prevents the proper evalua-
tion of the health consequences of smoking.
2 Defining Health Biases
In this section, we define absolute and relative health biases. Given the cross-sectional data
used for the empirical exercise, we consider a static model in which, depending on health
perceptions, an agent chooses risky health behavior that negatively affects her health. In the
empirical section, we will investigate how such biases relate to risky health behaviors.
Before we begin, we would like to reiterate that this paper remains agnostic about the
sources of biased beliefs. The origins of biased beliefs are still poorly understood. They have
been linked to image motivation (B´
enabou and Tirole,2002) and humans’ desire of being per-
ceived positively by others (Burks et al.,2013;Goette et al.,2015;Charness et al.,2018), as op-
posed to managing a favorable self-image (Santos-Pinto and Sobel,2005;K˝
oszegi,2006;Wein-
berg,2006) or self-serving biases (Babcock and Loewenstein,1997;Di Tella et al.,2015). For
example, Benoˆ
ıt and Dubra (2011) show that overconfidence—defined as overplacement—can
result from a rational Bayesian updating process. In addition, conformism to social norms or
“state-dependent reporting bias” may lead to systematic response error in self-reported health
(c.f. Kerkhofs and Lindeboom,1995;Lindeboom and van Doorslaer,2004). In this paper, we
neither study the origins of biased beliefs, nor do we make claims about the rationality of such
beliefs.
Rather, consider a risky health behavior xiwhich impairs individual health Hiaccording to
the health production function Hi=g(xi), with g0<0 and g00 >0. For each individual i, we
define the individual’s perceived health status e
Hias
e
HiAiHi=Aig(xi). (1)
The term Aimeasures the individual’s absolute health perception bias. If Ai=1, the individual
has a correct perception of her own health ( e
Hi=Hi), and a correct perception of the impact
of her risky health behavior on health ( e
Hi=g(xi)). If Ai>1, individual idisplays a positive
absolute health bias. This could be the case because she believes that her health is better than it
really is, or because she believes that her risky health behavior is less health-damaging than
it actually is. In either case, for a given health behavior, perceived health is better than true
5
health. If instead A<1, individual idisplays a negative absolute health bias. Our notion of an
absolute health perception bias focuses on the uncertainty about one’s own health condition
whose sources could be, for example, optimism or pessimism biases, self-deception, a lack of
health knowledge or reference group-dependent reporting biases. However, it is different from
uncertainty about future health consequences (which in our static notation would be captured
by uncertainty in g(xi)).3
Next, given Hi, define the objective relative position (the “ranking”) in the population
health distribution as:
ri100 ·F(Hi), (2)
where F(Hi)RHi
0dF(H)is the cumulative distribution function of population health (it
is multiplied by 100 to provide a ranking on a 0–100 scale, as in the survey question that elicits
this measure). Analogously, let the subjective health ranking be
e
ri100 ·PiF(e
Hi), (3)
where Piis the bias in the perception of the cumulative distribution function of population
health.4If Pi>1, individual ibelieves that there are more people in bad health than there
actually are; this can also be interpreted as pessimism bias about population health.
The comparison between ri=ri(Hi)and e
ri=e
ri(Pi,AiHi)allows us to identify the drivers
of health biases. Both rankings depend on risky health behavior and objective health. How-
ever, the subjective ranking e
ri, also depends on the absolute health perception bias Ai, and the
perception of the population health distribution Pi.
Remark 1 For a given reference population:
3Self-serving biases can be considered assuming that perceived health is e
HiAiHi=Aig(xi,Bi), where Biis
a measure of the self-serving bias and gB>0. This is the case, for example, when individuals underestimate the
health consequences of risky behavior because they believe that risky behaviors are less detrimental to health than
they actually are. Since both Aand Bincrease perceived health for a given risky behavior, henceforth, we assume
that self-serving bias B is absent, and we focus on the effect of absolute and relative health biases.
4Note that the precise wording of our survey question to elicit e
riis designed to minimize what has been iden-
tified and called “state-depending reporting bias” in SAH (Lindeboom and van Doorslaer,2004;Etil´
e and Milcent,
2006;J¨
urges,2007;Bago d’Uva et al.,2008;Ziebarth and Karlsson,2010). That is, the literature finds that survey
respondents have an implicit age-dependent reference group in mind when responding to the standard SAH ques-
tion; for example, older people tend to rate their health relatively better than younger people on the SAH scale. In
our survey question to elicit e
ri, we therefore specify “Imagine one would randomly select 100 people in your age,
[...].”
6
The objective health ranking ridecreases with risky health behavior and increases with better ob-
jective health: ri
xi
<0, ri
Hi
>0
The subjective health ranking e
ridecreases with risky health behavior and increases with better
objective health, the absolute health perception bias, and the population health pessimism bias:
e
ri
xi
<0, e
ri
Hi,e
ri
Ai,e
ri
Pi
>0.
For each individual i, we define the relative health perception bias Rias the difference between
the subjective and the objective ranking in the population health distribution:
Ri=e
riri(4)
We say that individual idisplays a positive relative health bias when Ri>0, and that she
displays a negative relative health bias when Ri<0. Manipulating equation (4) allows us to
decompose the relative health bias into two components:
Ri=100 [PiF(AiHi)F(Hi)](5)
=100 [F(AiHi)F(Hi)]+100 (Pi1)F(AiHi)
The first component of equation (6) depends on the absolute health bias Ai: it is equal to
zero if Ai=1, and it increases as Aiincreases.5The second component is equal to zero if
the perception of the population health distribution is correct (Pi=1), and it increases as Pi
increases.
Based on equation (5), we can state the following
Remark 2 For a given reference population and for a given belief Piabout population health, there exists
a one-to-one positive correlation between the absolute and the relative health perception bias.
The above remark emphasizes a convenient property that allows us to use the relative health
perception bias Rias a proxy for the absolute health perception bias Ai. In the following, we assume
that this property holds.
5In principle, the beliefs about population health could depend on the individual’s perceived health, i.e., Pi=
Pi(AiHi). In such a case, based on equation (5), higher perceived health AiHiincreases the relative perception bias
if P0(AiHi)F(AiHi) + P(AiHi)F0(AiHi)>0. As F(AH),F0(AH)and P(AH)are positive, the condition always
holds when P is constant (Remark 2), and also when P0(AiHi)>0.
7
3 Model
To investigate the role of perceived health for health behavior, consider an individual that de-
rives utility from risky health behavior x, perceived health e
H=AH =Ag (x), and consump-
tion of a numeraire good q. To fix ideas, consider the following utility function
V(x,e
H,q) = B(x)+H(e
H) + q(6)
The first term B(x)is the utility of a risky health behavior such as smoking, overeating, not
exercising or not sleeping enough. The second term H(e
H)describes the benefits of perceived
health. It can include mental and physical health, as well as any motive that links perceived
health to, e.g., social image concerns, peer pressure, conformism to social norms or self-esteem.
We assume that both, the utility of risky health behavior and the utility of health, are increasing
and concave.
Given income Mand prices p, the individual chooses risky behavior xby maximizing equa-
tion (6) subject to M=px +qand e
H=Ag (x), with x[0, M/p]. Replacing the constraints in
the objective function, this is equivalent to maximize U(x), where
U(x) = B(x) + H(Ag(x))+Mpx (7)
For an internal solution, the optimal amount of risky behavior xsatisfies the following first
order condition:
x:U0(x) = B0(x)p+AH0(Ag(x))g0(x) = 0 (8)
Equation (8) implies that optimal risky behavior trades-off the marginal benefits B0>0 of
risky behavior, its price p, and its marginal impact on perceived health AH0g0<0. The latter
term has a negative sign because more risky behavior reduces health (g0<0), with a magnitude
that depends on the absolute health perception bias A. In fact, the absolute health bias Aplays
a double role for optimal risky behavior. First, it affects perceived health ( e
H=Ag(x)) in the
marginal perceived health function H0(·). Because changes in Aaffect perceived health at the
margin, that is, the “last unit” of perceived health, we will call it the extensive margin of perceived
health. Second, Aaffects the magnitude of the marginal cost of risky behavior (H0g0). Because
8
this channel determines the impact of Afor all “inframarginal units” of risky behavior, we will
call it the intensive margin of perceived health.6
Ultimately, we want to assess how an increase in the health perception bias A(or its proxy
R, as per Remark 2) affects risky health behavior. This amounts to studying, omitting the
arguments,
dx
dA =g0
U00 e
HH00 g0
U00 H0(9)
The first and the second term on the right hand side describe the extensive and the intensive
margin of perceived health, respectively. The extensive margin implies a positive effect of the
health perception bias on optimal risky behavior (g0e
HH00/U00 >0). To give an example, the
effect on the extensive margin is akin to saying “Because (I think) I am healthy enough to not
exercise, I decide to not exercise.” In contrast, the intensive margin implies a negative effect
of the health perception bias on optimal risky behavior (g0H0/U00 <0). It is akin to saying
“Because I don’t want to jeopardize my (perceived) high health status, I will exercise more.”
Based on equation (9), the following holds:
Remark 3 When the absolute health perception bias increases, optimal risky behavior increases if the
impact of the absolute health perception bias on perceived health is larger than the impact on the marginal
costs of risky behavior, and it decreases otherwise.
The above remark shows that the balance between the extensive margin and the intensive
margin of perceived health determines whether optimal risky health behavior ultimately in-
creases or decreases with health perception biases. Recall that the extensive margin refers to
the effect of perceived health on risky behavior: as the agent thinks she is very healthy, she be-
lieves she can afford engaging in risky behavior. Accordingly, the better the perceived health,
the more she engages in risky behavior, which implies they they are complements. The inten-
sive margin, instead, implies that the absolute health bias increases the marginal costs of risky
behavior. Accordingly, the agent reduces her risky health behavior the larger her bias, which
implies that they are substitutes. Overall, when the health perception bias dominates the costs
of risky behavior (that is, when the extensive margin dominates), a larger health bias results in
more risky behavior.
6Our notions of extensive and intensive margin of perceived health are inspired by, but different from the notions
of extensive and intensive margin used to distinguish work participation and work effort.
9
Alternatively, we can describe equation (9) and the remark in terms of relative risk aver-
sion. Define σ e
HH00/H0>0 as the coefficient of relative risk aversion of perceived health,
denote α≡ H0g0/U00 >0, and replace in equation (9) to obtain dx/dA =α(σ1). Hence the
sign of dx/dA depends on whether the relative risk aversion of perceived health is larger or
smaller than 1. In the special case of a CRRA perceived health function, H(e
H) = 1
1σe
H1σ, the
coefficient of relative risk aversion σis constant and independent of perceived health. Accord-
ingly, risky behavior monotonically increases with the health perception biases if σ>1, and
it monotonically decreases if σ<1. If instead σ=1, the effect of Aon xis nil. In general,
however, the effect can be non-monotonic. For example, if the health function is quadratic or
exponential, the relation between health perception biases and health behavior is U-shaped.7
4 Datasets
To empirically identify the relationship between health perception biases and risky health be-
haviors, we exploit three representative German surveys. The first survey is the German Na-
tional Health Survey East-West 1991 (GNHSEW91s), which is representative of the German
population. The GNHSEW includes objective health measures, taken by health care profes-
sionals. By comparing specific clinical diagnoses of high cholesterol and high blood pressure
to respondents’ perceptions about such conditions, we generate measures of absolute health
perception biases.
The second survey is the Berlin Aging Study II (BASE-II), which is representative of elderly
residents in Berlin. This dataset includes objective health measures, taken by health care profes-
sionals, cognitive tests and rich socio-demographic background information. The BASE-II also
contains subjective health measures and the continuous and generic health measure SF12. Most
importantly, for this paper, we added a question to BASE-II that elicits respondents’ perceived
rank e
riin the population health distribution.
The third survey is the Innovation Panel of the representative German Socio-Economic
Panel Study (SOEP-IP). The SOEP-IP includes the standard socio-demographics of the SOEP,
self-reported health measures as well as the SF12. In addition to adding it to BASE-II, we also
included the same measure on respondents’ perceived rank in the population health distribu-
7For simplicity, in this example, Adoes not affect the utility of risky health behavior. In the more general case
U(x,e
H) + qwhere risky behavior and perceived health are non-separable, the sign of dx/dA depends on whether
the marginal utility of perceived health Ue
His large enough, as in the separable case. Specifically: dx/dA >0 if
and only if Ue
H>e
HU e
He
HgU e
Hx /g0
10
tion to SOEP-IP. Hence we can measure perceived ranking e
riand, together with the objective
ranking rifrom the SF12, generate measures of relative health perception biases, Ri.
4.1 German National Health Survey East-West 1991 (GNHSEW91)
The GNHSEW91 is a representative cross-sectional survey of the German population. It was in
the field in East and West Germany between 1990 and 1992 (Robert Koch Institut,2012). Many
questions are nutrition and health-related; our GNHSEW91 working sample consists of 6,429
respondents. Importantly, health care professionals measured both, the clinical blood pressure
as well as the cholesterol levels of all respondents (Panel B of Table A1, Appendix). In addition,
before the clinical examination by nurses and physicians, GNHSEW91 surveys respondents’
perceptions about whether they have high blood pressure or high cholesterol levels in a self-
completed questionnaire (Panel C of Table A1, Appendix).8Comparing perceived ( e
Hi) and
true health (Hi) allows us to measure absolute health perception biases (Panel A of Table A1
and Section 5.1).
Health Behavior. Panel D of Table A1 (Appendix) lists measures of risky health behaviors.
Thirty-three percent are current smokers and 49% do not exercise at all (No sports). Thirteen
percent consume alcohol daily and the average BMI is 26.6.
4.2 Berlin Aging Study II (BASE-II)
The BASE-II consists of several parts: The first part is the Socio-Economic Module that com-
prises standard questions of the Socio-Economic Panel Study (SOEP) (Wagner et al.,2007). This
part includes self-reported socio-demographics, health and health behavior measures (B¨
ockenhoff
et al.,2013).
The second part is the Clinical Module. It includes clinical health measures, taken in the
Charit´
e University Hospital of Berlin. Additional parts include cognitive and other tests which
were administered by psychologists but which are not the focus of this paper.9The BASE-
II is representative of the elderly Berlin population up to age 89. As a supplement, BASE-
II also surveys a sample of younger Berlin residents aged 18 and above; the ratio between
respondents above and below 60 is 3:1 (see Appendix, Table A2). Bertram et al. (2014) provide
8Robert Koch Institut (1995) describes the exact protocol and the order of the examination. For example, after
the blood samples were taken by a registered physician, the serum was analyzed in a lab. The findings were
summarized and commented and sent by mail to the survey participants.
9Our findings are robust to controlling for cognitive measures. Detailed results are available upon request.
11
more information on the BASE-II. Our working sample consists of 1,804 respondents without
missings on relevant variables.
For this paper, we included a measure to elicit e
riin the Socio-Economic Module of BASE-II,
which was in the field between September and December 2012. Section 5.2 discusses the health
bias measures in detail, also see Panel A of Table A2.
Health Behavior. Panel B of Table A2 lists measures of risky health behaviors: smoker,no
sports,unhealthy diet,obese and BMI. As seen, 12% smoke, 36% do not exercise, 39% have an
unhealthy diet, and 13% are obese.
Socio-Demographics. Panel C of Table A2 lists socio-demographic control variables. There
are five main categories: (i) Demographics, (ii) Education, (iii) Employment, (iv) Risk Aver-
sion, and (v) Big-Five personality traits (openness, conscientiousness, extraversion, neuroti-
cism, agreeableness). The average age of BASE-II respondents is 60; slightly more than half of
them are female and married; a quarter are single. Fifty-six percent of the sample finished high
school (13 school years) and almost half are still full-time employed. In our regression mod-
els, we also control for risk aversion and trust, which are important covariates when eliciting
subjective beliefs (e.g. Harrison et al.,2015,2017). Accordingly, 15% of BASE-II respondents
are risk loving (highest three categories of the standard 0 to 10 Likert risk aversion scale, see
Dohmen et al.,2011). Finally, we also control for the Big-Five. The five dimensions are simple
averages over three or four subscales which range from one to seven (Richter et al.,2013). Con-
scientiousness has the highest average of 5.6 and Agreeableness the lowest with 3.8 (Panel C,
Table A2).
4.3 Socio-Economic Panel Study – Innovation Panel (SOEP-IP)
Since 2012, the SOEP has been inviting researchers to submit proposals for innovative survey
questions (Richter and Schupp,2015). Proposals are then reviewed by an expert committee. If
accepted, the questions become part of SOEP-IP, which is in the field annually from September
to December. SOEP-IP respondents also answer the regular SOEP core questions (Richter and
Schupp,2017). In 2014, a total of 1,377 respondents answered the same health perception mea-
sure, e
r, that we also included in BASE-II. Comparing relative health perceptions to true health
allows us to construct relative health perception bias measures Ri(see Panel A of Table A3 and
Section 5.2).
12
Health Behavior. In 2014, the SOEP-IP did not ask about smoking, exercising, and respon-
dents’ diet. However, the SOEP-IP elicited the average hours of sleep (Richter and Schupp,
2015). On average, Germans sleep 6.8 hours during the week and 7.6 hours on weekends
(Panel B of Table A3). We use these information to generate sleep gap measures that indicate
the difference to eight hours of sleep.
Socio-Demographics. As above, Panel C of Table A3 lists socio-demographics. Because
BASE-II and SOEP-IP both include SOEP’s socio-demographic core questions, we generate
almost identical socio-demographic control variables. By design, representative SOEP-IP re-
spondents are younger (51 vs. 60 years) but the shares of female and married respondents are
very similar, slightly above fifty percent. In SOEP-IP, two thirds are full-time employed and the
average monthly net income is e1,768.
5 Measuring Health Perception Biases
This section shows how we operationalize our measurement of health perception biases using
survey data.
5.1 Measuring Absolute Health Perception Biases
To measure absolute health perception biases, we use the German National Health Survey East-
West 1991 (GNHSEW91). This dataset contains information on individual blood pressure and
cholesterol levels, collected by the Institute for Prevention and Public Health in Berlin, Ger-
many. We use these measures as proxies for the individual objective health status Hi. We di-
chotomize these continuous objective measures depending on whether they are above or below
the medically defined threshold to indicate specific clinical conditions. Specifically, for blood
pressure, we define that a respondent has high blood pressure, and BPiequals one, if the sys-
tolic value is larger than 160 mmHg and/or the diastolic value exceeds 95 mmHg; BPiequals
zero otherwise. Analogously, for high cholesterol levels, we define a dummy Choliequal to
one for values larger than 6.2 mmol/l, and zero otherwise. Because GNHSEW91 also elicits
perceptions about these conditions, we then compare these objective clinical outcomes BPiand
Choliwith respondents’ perceived high blood pressure, g
BPi, and high cholesterol levels, g
Choli,
13
to obtain an assessment of absolute health bias. Each surveyed individual provides self-assessed
binary measures of g
BPiand g
Choli.10
Panel B of Table A1 (Appendix) shows a mean cholesterol level of 6.1 millimole per liter
(mmol/l) and that 44% of all Germans have high cholesterol levels (BPi=1). Panel B of Table
A1 also shows mean systolic blood pressure levels of 135 millimetres of mercury (mmHg) and
mean diastolic blood pressure levels of 83 mmHg.11 Following the official WHO definition at
the time, 21% of Germans had hypertension (Choli=1).12 Panel C shows that 21% of respon-
dents knew that they suffered of hypertension (g
BPi=1) and that 25% knew that they had high
cholesterol ( g
Choli=1).
[Insert Figure 1about here]
As good health corresponds to BPi=0 or Choli=0, positive health biases results if
g
BPi<BPior g
Choli<Choli. Figure 1a illustrates the four possible combinations of the binary
measures of objective and perceived health for cholesterol. Individuals in the bottom-right cor-
ner display positive absolute health bias. This corresponds to 30% of respondent who actually
have high cholesterol levels, but are not aware of it. The bottom-left corner shows that 50%
of all respondents do not have high cholesterol levels and consistently report that they do not
have high cholesterol. As shown in the top-right corner, 14% correctly state that they have high
blood cholesterol levels. Henceforth, we ignore the 6.5% in the top-left corner of Figure 1a who
had no high cholesterol at the time of the survey, but who claim that they have been diagnosed.
The reason is that we cannot accurately assign these respondents to either being accurate or be-
ing pessimistic about their health. This is because the language of the German question asks
whether respondents have ever been diagnosed with high cholesterol levels. It could well be
that respondents had high cholesterol levels in the past (which is why the respondents an-
swered with “yes”) but have changed their lifestyle and do not have high cholesterol levels at
the time of the survey (which is why the clinical measurement yielded no such indication).
Figure 1b has the same setup and shows the analogous distribution for high blood pres-
sure. Accordingly, 66% of all respondents correctly state that they do not have high blood
10Note that there is a small literature on misreporting of clinical diagnoses (Baker et al.,2004;Davillas and Pud-
ney,2017;Choi and Cawley,2018). It is certainly up to scientific debate on how to define this phenomenon. People
are either unaware and have biased health perceptions (our interpretation) or they are aware of their health condi-
tion but deliberately misreport it, for example, due to a desirability bias.
11Each measure was taken three times from each respondent; we use data from the second measure.
12In the meantime, the official definitions have been downgraded. In November 2017, the American Heart Asso-
ciation and the American College of Cardiology redefined the thresholds to 130/80 (American Heart Association,
2017).
14
pressure (bottom-left corner), and 12% correctly state that they do have high blood pressure
(top-right corner). Nine percent indicate that they never had high blood pressure although
the clinical measures show the opposite (bottom-right corner).13 The smaller perception bias
for high blood pressure as compared to high cholesterol is consistent with the notion that it is
easier to check for high blood pressure than high cholesterol levels outside of clinical settings.
Following the arguments above, we also ignore the 13.5% of respondents in the top-left corner
of Figure 1b.
Finally, we would like to comment on the fact that the GNHSEW91 is already three decades
old. One may hypothesize that health knowledge was not as advanced at the time as it is
today. There is certainly reason to believe that the average person’s—and also science’s—
understanding on the negative health effects of high blood cholesterol and high blood pressure
was not as advanced as it is today, but they were clearly known (cf. Glanz,1988;Weiss,1972).
Moreover, it was known already at the time that smoking, obesity and heavy alcohol consump-
tion are detrimental to health while exercising has a positive impact (cf. Feinleib,1985;Institute
of Medicine,1990;Trichopoulos et al.,1981).
What’s more, the GNHSEW91 was in the field shortly after the German Reunification of
1990 and it was precisely a purpose of this survey to produce a representative picture of the
health and diet of East and West Germans. Consequently, Figure A1 (Appendix) plots the
shares of respondents with health perception biases regarding high blood cholesterol and high
blood pressure levels separately for East and West Germans. As seen, the shares of East Ger-
mans who were not aware of their high blood cholesterol (39%) and high blood pressure (14%)
is significantly higher than the shares of East Germans who were not aware of their high blood
cholesterol (25%) and high blood pressure (7%). As an unbalanced diet is one main risk factor
for high blood cholesterol and high blood pressure, this in line with research showing that,
after the fall of the Wall, East Germans consumed a significant amount of novel Western food
when it became readily accessible (Dragone and Ziebarth,2017).
5.2 Measuring Relative Health Perception Biases Ri
To measure relative health perception biases Ri=e
riri, we make use of the BASE-II and the
SOEP-IP.
13Note that the prevalence of the medical condition also determines the prevalence of the absolute health bias.
However, correcting responses by the prevalence rate is outside the scope of this paper.
15
To measure Hi(which is needed to construct ri), both the BASE-II and the SOEP-IP contain
the standard SAH measure as well as the SF12 measure. Both measures have been routinely
used by health economists and public health scientists. SAH asks about the overall health sta-
tus; respondents self-categorize as being in excellent, very good, good, fair, or poor health.
However, although widely available and easy to collect, the literature has documented system-
atic SAH response biases with respect to age and gender (Lindeboom and van Doorslaer,2004;
J¨
urges,2008;Bago d’Uva et al.,2008;Ziebarth,2010;Spitzer and Weber,2019), which we control
for in our regressions.
However, to minimize concerns about reporting biases, in our main specifications, we em-
ploy the generic and continuous SF12 as Himeasure (Andersen et al.,2007). The SF12 belongs
to the “health-related quality of life measures.” Using a specific algorithm, the SF12 weights
and aggregates the answers to twelve health questions into a physical health (pcs) and a mental
health (mcs) summary scale. Compared to SAH, the SF12 is a “more” objective health measure
and was developed to minimize reporting biases. It “can be used to compare the health of
different groups, for example, the young and the old or the sick and the well” (RAND,1995).
Both subscales of the SF12, pcs and mcs, have continuous values between 0 and 100, mean 50,
and a standard deviation of 10. We use equal weights of 0.5 to generate the overall continuous
SF12 measure. In a robustness check, we also use the physical and mental health components
separately as the benchmark Himeasure. Figures A2 and Figure A3 in the Appendix show the
distributions of SAH and SF12 in our BASE-II (Figure A2) and SOEP-IP sample (Figure A3).
The left panels refer to SAH and the right panels refer to SF12. The health distributions appear
very similar, both across measures and across databases.
SAH and SF12 allow us to infer Hias well as the population health distribution F(Hi).
Then we calculate the individual rank riin the health distribution. For SAH, each respondent
self-categorizes into one of the five SAH categories. We assign every respondent the upper cdf
threshold of the category chosen in the SAH distribution. For example, 9% of all respondents
are in the highest category “excellent” health. Hence, we assign ri,S AH =91 to all respondents
in the second highest category “very good” and, using the same principle, we do the same
for the other categories. Because SF12 is continuous, ranges from 0 to 100 and has mean 50, it
directly yields riwithout further manipulation.
To measure e
ri, we added the following question to BASE-II and SOEP-IP: “Imagine one
would randomly select 100 German residents in your age, what do you think: How many of
16
those 100 people would be in better health than you?”.14 From the raw untransformed response
to this question, e
bi, we compute e
ri=100 e
bi.
In both surveys, we obtain high response rates of above 90% for our e
bimeasure. Even
among the elderly in BASE-II, only 10 respondents (<1%) are coded “don’t know” and 129
respondents (6%) are coded “does not apply.” The high response rates may be a function of
the natural reference group—100 German residents in the same age group. This framing al-
lows meaningful comparisons without being too restrictive or too complex. Note that, despite
avoiding many of the methodological criticisms of earlier studies (Benoˆ
ıt and Dubra,2011), our
question does not elicit entire belief distributions (Di Girolamo et al.,2015). Moreover, we did
not specifically incentivize respondents (Harrison and Rutstr¨
om,2006). Eliciting entire belief
distributions in an incentive-compatible environment is typically feasible in lab experiments
(Harrison,2015), which is costly and can only be implemented in large samples under specific
conditions. In addition to the advantages above, maybe the main advantage of our measure
of e
biis its simplicity and cost-effectiveness. Using one simple question, our proposed question
has the power to elicit subjective relative beliefs in representative population surveys.
[Insert Figure 2about here]
Figure 2shows the raw untransformed distributions of e
bifor BASE-II and SOEP-IP. Un-
der the assumption of 100 random German residents being orderly ranked from 1 to 100 and
under full rationality and common priors, e
biwould be uniformly distributed between 0 and
100 with a mean of 50 (Goette et al.,2015). However, as seen in Figure 2, few respondents say
that more than 50 respondents in their age would be in better health and the distribution is
clearly skewed to the left. It is worthwhile to emphasize the similarity of the e
bidistributions in
BASE-II and SOEP-IP; the mass of the distributions lies between 10 and 30. In other words, a
significant share of respondents believe that (only) 10-30 out of 100 people are in better health
(e
bi(10; 30)) implying that they rank themselves in the 70th to 90th percentile of the popu-
lation health distribution, e
ri(70; 90). This yields first evidence for the existence of health
overconfidence at the population level.
[Insert Figure 3about here]
14This survey question has been successfully tested in other contexts. For example, using the same format, respon-
dents in the Swiss “Amphiro” study were asked about their income position, their water use, and their knowledge
of energy conservation (Tiefenbeck et al.,2018;Friehe and Pannenberg,2019).
17
More evidence for the existence of relative health perception biases is illustrated in Figure 3,
which plots the bins of e
rion the x-axis and the average values for rion the y-axis. The scatters,
whose size indicate the share of respondents falling into each bin, would be lined up along the
45-degree line if e
ri=ri. However, as seen, while the scattered line has a slightly positive slope,
it is clearly flatter than the 45-degree line. Again, the similarity between BASE-II and SOEP-IP
is worthwhile to emphasize. Moreover, the size of the scatters reflect the mass of the perceived
health rank distributions which fall into the 70th to 90th percentile bins, whereas the true health
status of respondents who believe that only 10 to 30 randomly selected people would be in
better health is only about average.
Using riand e
ri, we can now calculate Rifor each respondent in BASE-II and SOEP-IP. Be-
cause BASE-II is representative of the elderly Berlin population, whereas SOEP-IP is represen-
tative of the entire German population, comparing the results of both surveys will inform us
about the generalizability of the empirical findings. Calculating Ri,SF12 =e
riri,SF12 is straight-
forward because both the SF12 and e
riare continuous. When calculating Ri,SAH =e
riri,SA H ,
recall that e
riis continuous, but SAH has five categories. However, because e
riis continuous and
varies within SAH categories, Ri,S AH is continuous as well, as shown by Figures 4a (SOEP-IP)
and A4a (BASE-II). Both figures also demonstrate that Ri,SAH looks very similar in BASE-II and
SOEP-IP and that the gender differences are negligible.
[Insert Figure 4about here]
Figures 4b (SOEP-IP) and A4b (BASE-II) show the distributions of Ri,SF12. As seen in Panels
A of Tables A2 and A3, the mean Ri,S F12 values are 14 (SOEP-IP) and 20 (BASE-II), that is,
clearly positive and implying that respondents overestimate their ranks on average by 14 and
20 positions. Conditional on having a positive absolute health bias, respondents overestimate
their health rank by 23 (BASE-II) and 19 (SOEP-IP) positions. Again, the Ri,SF12 distributions
are very similar for BASE-II and SOEP-IP.
Table A4 (Appendix) shows determinants of Ri,SF12 using the representative SOEP-IP. As
seen, women are more likely to be positively biased, as are childless respondents. Interestingly,
educational and job characteristics as well as risk tolerance levels are not significant predic-
tors of positive health perception biases. The fact that education is no significant predictor of
the biases suggests that it is not health knowledge related to formal education that is a main
18
underlying mechanism in this setting.15 Interestingly, female and being childless also predict
negative biases significantly and carry the same sign, implying that women and childless re-
spondents are less likely to have negative biases, as are white collar workers. Finally, when
investigating the Big-Five measures as predictors of health perception biases using the BASE-
II, we find that that conscientiousness, extraversion and agreeableness are negatively related to
positive and negative health biases, whereas neuroticism and openness are stronger and highly
significant predictors of health overconfidence (see Table A5 in the Appendix).
6 Health Perception Biases and Risky Health Behaviors
In this section, we first study the empirical link between absolute health perception biases and
risky health behaviors using the GNHSEW91. Then, we study the link between relative health
perception biases and risky health behaviors using the BASE-II and SOEP-IP. We will provide
non-parametric evidence and evidence from multivariate regression models.
6.1 Absolute Health Perception Biases and Risky Health Behaviors
Figure 5tests whether respondents who have biased perceptions about their blood cholesterol
levels are more likely to (a) not exercise, (b) have higher BMIs, (c) drink alcohol daily, (d) smoke.
Figure 6tests the same relationships for respondents who have biased perceptions about their
blood pressure levels, see Section 5for details about how we generate the perception bias mea-
sures. Each of the figures shows four bar diagrams along with 95% confidence intervals.
[Insert Figures 5and 6about here]
Figure 5a shows that respondents who state that they do not have high cholesterol but
who, in fact, do have high cholesterol are a highly significant 11 percentage points (ppt) more
likely (43% vs. 54%) to not exercise at all. The BMI differential is also significant (Figure 5b).
Similarly, respondents with absolute health perception biases are significantly more likely to
drink alcohol daily—the share of daily drinkers is almost 50% higher among this group (21%
15In line with that finding, the R2only increases very slightly by 0.003 when adding education controls separately
to our main regressions and the size and significance of the main regressors remain unchanged. On the other hand,
there is some evidence that formal education is a significant predictor for biases in BASE-II (see Table A5); however,
this database is only representative for elderly people in Berlin.
19
vs. 14%, Figure 5c).16 Figure 6d, however, does not provide much evidence that smoking is
significantly linked to biased perceptions about high cholesterol levels.
Comparing Figure 6to Figure 5, the similarity and robustness of the link between both ab-
solute health perception bias measures and four risky health behavior measures is worthwhile
to point out. Not only do all statistical links have identical signs and significance levels, but the
risky behavior differentials and their sizes are also very similar. This is even more surprising,
given the low correlation between the two perception bias measures of only 0.11.17
In conclusion, there is robust evidence that absolute health perception biases are signif-
icantly linked to three out of four risky health behaviors. According to Proposition 3, this
implies that the extensive margin effect dominates the intensive margin effect meaning that
positive health biases induce people to engage in more risky behavior because they (wrongly)
believe that they can “afford” it. One exception appears to be smoking, where the intensive
margin effect appears to be stronger. This intensive margin effect lowers the inclination to
engage in risky behavior because the bias increases the marginal costs of risky behavior, see
Section 3for more details.
6.2 Relative Health Perception Biases and Risky Health Behavior
Figure 7non-parametrically links Rto xacross the entire Rdistribution using kernel-weighted
local polynomial smoothing plots. Table 1provides analogous parametric multivariate regres-
sions using a rich set of controls.
[Insert Figure 7about here]
Figure 7a shows a monotonically increasing relationship between positive relative health
biases, R>0, and not exercising. On average, 30% of those who accurately assess, or who
underestimate their health, do not exercise at all. This share monotonically increases to 50%
for respondents who overestimate their rank in the population distribution by 50 ranks; that is,
who exhibit a strong positive health bias.
16While moderate drinking has been linked to health and labor market benefits (Renaud et al.,1999;Ziebarth and
Grabka,2009;Holst et al.,2017), heavy drinking has detrimental health effects and national guidelines recommend
to abstain from drinking two days per week (Rehm et al.,2001;National Health Service,2012).
17An attentive referee suggested to check for information about medication intake to lower the blood pressure
or high cholesterol levels. Indeed, the GNHSEW91 does collect this information in self-reports before the clinical
examination. However, as expected, only 6 out of 1,908 respondents who say that they don’t have high cholesterol
and 9 out of 600 respondents who say that they don’t have high blood pressure indicate that they take medications
against that disease at the same time. The results are robust to dropping these respondents.
20
Next, we run the following parametric regression model controlling for a rich set of socio-
demographics:
xi=β0+β1R+
i+β2R
i+Ziβ3+ρt+ei(10)
where xirepresents risky health behavior and Ristands for our measure of relative health
perception bias. Specifically, we will replace the continuous Rimeasure with two measures
R+
i{Ri|Ri(0; 100)and R
i{Ri|Ri(100; 0).R+
iis truncated from below and measures
the degree of positive health bias. R
iis truncated from above and measures the degree of
negative health bias. Below we provide robustness checks on the appropriateness of this spline
which allows us to separately study health overconfidence and health underconfidence.
Zicontains socio-demographic controls as listed in the descriptive statistics in the Ap-
pendix. Moreover, we include interview month fixed effects, ρt;eiis the error term.
[Insert Table 1about here]
Tables 1and 2show our main results using Ribased on the SF12 benchmark and 10 regres-
sion models as in equation (10).18 The equivalent table using Ribased on the SAH benchmark
is in Table A6 (Appendix). The first two columns of Table 1show the results for not exercis-
ing as an outcome, the next pairs of columns document the results for being obese, following
an unhealthy diet, and being a smoker, respectively. Table 2shows results for sleeping less
than 8 hours during the week. While the odd-numbered columns solely control for socio-
demographics, education and interview month fixed effects, the even-numbered columns ad-
ditionally control for employment characteristics and income.
No Sports. Beginning with the outcome no sports in the first two columns of Tables 1and
A6, the results confirm the non-parametric findings in Figure 7: there is no evidence that a
negative health perception bias is significantly linked to not exercising. By contrast, we find
a highly significant link between Ri>0 and not exercising: an increase in Riby 10 ranks is
associated with a 2.2ppt higher likelihood to not exercise (column (2), Table 1). The size of
the association is larger when using RSA H (columns (1) and (2), Table A6) but overall robust.
Also note the very consistent evidence for the same outcome when using absolute health bias
measures in Figures 5a and 6a.
18In Table A7 in the Appendix, we replicate Tables 1using probit estimation instead of OLS. The results (marginal
effects) are almost identical.
21
BMI, Obesity and Unhealthy Diet. Figure 7b and columns (3) and (4) of Tables 1and A6
show the equivalent findings for BMI and obesity, while Figure 7c and columns (5) and (6) of
Tables 1and A6 show the findings for following an unhealthy diet. Both figures and all eight
regressions reinforce our previous findings.
Figure 7b shows a non-linear relationship between Ri>0 and BMI which is very similar to
Figure 7a: Respondents who accurately assess their health or who underestimate their health
do not have higher BMIs. However, the average BMI increases monotonically in the size of
the health bias for Ri>0. In other words, the more people overestimate their health, the
heavier they are. To test this link parametrically, we generate a binary obesity indicator as
dependent variable19 and run regressions similar to equation (10). The results in Tables 1and
A6 confirm the non-parametric visual evidence: An increase in the health bias Riby 10 ranks
is associated with a 1.2ppt (about 9%) higher probability to be obese in columns (3) and (4) of
Table 1; as above, the effect size is larger for RSAH but generally robust (Table A6, Appendix).
The inclusion of rich sets of individual control variables do not affect the size of the empirical
relationship.
Figure 7c and columns (5) and (6) of Tables 1and A6 corroborate the findings above.
Whereas no link exists for respondents with Ri0, that is, unbiased or pessimistic respon-
dents, we find a clear and positive statistical link between having a positive health bias and
eating unhealthy. An increase in the bias by 10 ranks increases the likelihood of an unhealthy
diet by 1.2ppt (or about 3.1%).
Smoking. Figure 7d and columns (7) and (8) of Tables 1and A6 show the results for being
a smoker. The graphical evidence provides no evidence for a statistical link between health
perception biases and smoking status. This is confirmed by the parametric regressions—none
of the health bias measures is significantly linked to smoking status and the effect sizes are
very small.20 This finding is in line with the findings from above in Figures 5and 6, where we
found no association between the absolute health bias and smoking. According to our model,
this implies that the health perception biases operate through a strong intensive margin effect,
which emphasizes the costs of smoking (Section 3). This finding is consistent with Darden
(2017), who reports that updated (and objective) cardiovascular biomarker information has
19Using the continuous BMI measure yields robust results.
20We also assessed the intensive margin of smoking, i.e., the number of cigarettes smoked by each smoker per
day. There as well we do not find any significant link and the effect sizes are very small (detailed results available
upon request).
22
not altered smoking behavior in the population of the Framingham Heart Study—Offspring
Cohort.
[Insert Figure 8and Table 2about here]
Sleep. Finally, using the representative SOEP, we study relative health perception biases
and sleep. Our outcome indicates the “sleep gap” between the actual hours of sleep and eight
hours.21 Figure 8a shows that the sleep gap is monotonically increasing in the size of the health
bias for most parts of the bias distribution from about -15 to +40. Interestingly, it looks like this
link weakens for both very high positive as well as negative biases such that we obtain what
looks like an inverse U-shaped pattern. Note that our model is flexible and powerful enough
to rationalize such non-linear pattern. Specifically, through the lens of the model, the pattern
would imply that the extensive margin in equation (9) dominates over the [-15;40] range of
support, whereas the intensive margin dominates outside this range among extreme positively
or negatively biased people. In Figure 8b, for sleep on weekends, we see again no clear as-
sociation for respondents with no or negative biases but a positive relationship among very
positively biased respondents.
Table 2shows the equivalent regression results using the sleep gap during the week as
dependent variable (results for the weekend are similar and available upon request). We find a
robust and clear link between a positive health perception bias and the likelihood to sleep fewer
hours. The findings in the first two columns even suggest that the association could carry over
for people with negative health biases, but with reversed signs; that is, those who are overly
pessimistic about their health may sleep significantly more than those who correctly assess
their health, implying that the intensive margin and thus the marginal costs of not sleeping
enough dominates here and negatively biased people do not want to jeopardize their health
through a lack of sleep (Section 3).
Robustness. In the following, we conduct a series of robustness checks. First, Table 3
replicates Table 1but additionally adds several risk aversion measures as well as all measures
of the Big-Five as listed in Table A2 (Appendix). As seen, the results are robust.
[Insert Tables 3and 4about here]
21There is heterogeneity in the hours of sleep that an individual needs but the large majority of people need
between 7 and 9 hours of sleep. However, a significant share of people in industrialized countries are permanently
sleep deprived, which has strong negative health consequences (Giuntella and Mazzonna,2019;Jin and Ziebarth,
2020).
23
Second, we examine the robustness of our findings with regard to the specification of spline
estimates that cut the spline at R=0 in equation (10). As we argue above, the distinction be-
tween overconfidence and underconfidence is theoretically meaningful. This makes the choice
of R=0 the natural starting point. We assess its robustness by asking whether there is a
break point for the spline that better fits the data than R=0. We do this by performing a grid
search over the interval R[15, 30](in steps of 0.1 points) to define the break point R, and
conditionally on R, estimate the remaining parameter by OLS. This allows us to estimate the
best-fitting spline to the data. Specifically, we test whether R=0 is rejected by the data, as
R=0 is nested within the class of models we estimate. Table A8 in the Appendix shows the
results. Most importantly, we cannot reject the null hypothesis of R=0, with the lowest p-
value being 0.19 (in column 1 of Table A8). The point estimates for the upper part of the spline
are unchanged, though their standard errors are larger. Further, we never find an estimate of
Rthat is significantly different from zero. Thus, we conclude that R=0 is an appropriate
specification for the spline.
Third, in addition to various possible interpretations of empirical measures of health biases
as well as the challenge to induce clean exogenous variation, there exists another structural
challenge in the literature: “ceiling effects.” Specifically, from a statistical standpoint, the better
the actual relative health status ri, the lower the probability that someone has a positive health
perception bias, by definition. Table 4shows a robustness check that trims the relative health
perception bias distribution and eliminates the top and bottom quintiles of Ri.22 Otherwise
the model is the same as in equation (10). The column headers of Table 4show the dependent
variables; the models in the uneven columns do not control for employment characteristics and
income, whereas the models in the even columns do.
Table 4also provides a robustness check for eliminating potential guesses of e
ri. Because it
could be that the peak at e
ri=50 in Figure 2represents respondents who did not know the
response and simply guessed.23 In addition to trimming the distribution, the sample used for
Table 4also omits these potentially guessing respondents. The result of the eight models in
Table 4corroborate our findings above: While there is no evidence that the likelihood to smoke
increases or decreases in the bias, the likelihood to engage in risky health behavior increases in
the bias for the other health behavior measures.
22The findings are robust to alternative ways to trim the distribution, e.g., trimming the top and bottom 30 per-
centiles. Those results are available upon request.
23Because of space restrictions in the surveys, we could not elicit the certainty with which respondents answered.
Future research should address this limitation.
24
Finally, Table A9 (Appendix) makes an attempt to differentiate the effects by physical and
mental health. This exercise comes with the limitation that our question to elicit e
riasks about
health in general, not specifically about physical and mental health. However, the SF12 allows
us to compute continuous measures of mental and physical health which we use as the bench-
mark in Table A9 (Appendix). As seen, in line with the prior that most people likely think about
physical health when answering our question to rank themselves in the population health dis-
tribution, the results suggest that our main findings are driven by physical health. Moreover,
most of the risky behaviors studied here are linked to negative physical health effects.
To summarize our empirical findings: First, except for the case of smoking, we find stable
and statistically significant links between health perception biases and risky health behaviors.
Second, no such links exist for respondents without bias or who exhibit a negative health per-
ception bias. Third, the findings are robust with respect to two self-reported general health
measures as well as two very specific objective physical health measures. Fourth, the findings
are also robust for four health behaviors: exercising, eating, drinking and sleeping. Fifth, the
empirical findings are consistent for both absolute and relative health perception biases. Lastly,
interpreted in the context of our model, except for smoking, the results imply that the exten-
sive margin effect of the bias dominates the intensive margin effect. In other words, people who
overestimate their health engage in a suboptimally high level of risky health behavior because
they wrongly believe that their bodies could tolerate such behavior.
7 Conclusion
This paper formally introduces the notion of health perception biases to the health economics
literature. It proposes an empirical quantification and discusses its prevalence using three dif-
ferent datasets from Germany. Then, it investigates theoretically and empirically whether and
how health perception biases could affect risky health behavior.
We consider two related notions, absolute and relative health perception biases. The relative
measure, which is based on the difference between the objective and the perceived rank in the
population health distribution, was explicitly elicited for the purpose of this study. One advan-
tage of this measures is the possibility to elicit it in large-scale surveys at relatively low costs.
Moreover, both measures provide comparable information as we show that, under plausible
conditions, there exists a one-to-one positive mapping between these two health perception
measures.
25
Using a simple model, we show that health perception biases affect individual health be-
havior at both the extensive and the intensive margin. These margins operate in opposite di-
rections. The extensive margin refers to the perceived health status, inducing an individual to
engage in more risky behavior because she believes she is healthy enough to “afford” it. The
intensive margin of health biases operates in the opposite direction: it inflates the perception
of the marginal costs of risky behavior, inducing an individual to refrain from it. In the for-
mer case, health biases and risky behavior are substitutes, while in the latter case, they are
complements.
Which effect dominates is an empirical question. We study it using three German datasets.
Our empirical findings are robust and consistent across the different health bias measures and
the three datasets. First, we show that health perception biases are pervasive in the health do-
main. For example, about 30% of all respondents of a representative German health survey
which includes blood sampling were unaware of their high cholesterol levels. Similarly, us-
ing two other representative surveys, we find that about 30% of all respondents overestimate
their position in the population health distribution by at least 30 ranks. We take these find-
ings as evidence that absolute and relative health perception biases exist systematically and are
widespread.
Second, we find that health perception biases are significantly linked to risky health behav-
iors, with the notable exception of smoking. We find that people who overestimate their health
are more likely to not exercise, to eat unhealthy, drink alcohol on a daily basis, and sleep fewer
hours. They also have significantly higher BMIs and are more likely to be obese. Interpreted
through the lenses of our model, these statistical findings are consistent with the notion that
the extensive margin of health perception biases dominates the intensive one. In other words,
people who overestimate their health engage in excessive (suboptimal) risky behavior because
they believe they are healthier than they really are.
However, third, among those who more accurately assess or underestimate their health, we
do not find much evidence that health perception biases are related to risky health behavior—a
result that is consistent with the intensive margin of perceived health offsetting the extensive
margin. Interestingly, smoking is not significantly correlated with health biases either. If future
research corroborates this finding, it could suggest that public health anti-smoking campaigns
aimed at altering smokers’ perceptions by emphasizing the health cost of smoking are less
26
effective than alternative policy tools such as, e.g., taxation and smoking bans (Viscusi,1990;
Becker et al.,1994;Chaloupka,1991;Gruber and K˝
oszegi,2001).
In terms of policy implications, our general results show that people with biased health per-
ceptions can be a fruitful target group for effective public health campaigns aimed at reducing
risky health behaviors. Adding regular health check-ups and screenings to the essential health
benefit packages, in addition to nudging people to seek regular feedback about their health,
can be a desirable policy. However, we also find that debiasing individuals’ health perceptions
would not only operate on the extensive margin of perceived health biases, but also affect the
intensive margin. Accordingly, people could possibly engage in more risky behavior because a
better than expected health assessment could reduce the fear of the negative consequences (that
is, the perceived marginal costs) of risky behavior. This result is particularly relevant when the
risky behavior produces negative externalities, because the underestimation of the marginal
cost to society adds to the underestimation of the individual costs.
For decades, a rich strand of research in health economics has theoretically and empirically
identified what researchers have labeled “reporting heterogeneity,” “state-dependent reporting
bias” or “scale of reference bias” (cf. Lindeboom and van Doorslaer,2004;Etil´
e and Milcent,
2006;J¨
urges,2007,2008;Bago d’Uva et al.,2008;Spitzer and Weber,2019). This paper shows
that “health perception biases” exist. However, it is beyond the scope of this paper to investi-
gate the sources of such biases. The literature discusses optimism biases, self-deception biases,
a lack of health knowledge or conformism to social norms as possible sources. An open ques-
tion for future research is whether and how such implicit biases could also affect self-reported
health measures and, in fact, be part of what has been previously identified as “reporting het-
erogeneity.”
Finally, our study is a first attempt to marry research on biased beliefs and risky health be-
havior, both from a theoretical and an empirical perspective. Although our empirical analysis
is solely based on statistical associations, we believe it is a fruitful avenue of research to study
the relevance of individual perceptions and beliefs for health behavior. In particular, future re-
search should investigate the generalizability of our results for different countries and different
institutional setups.
References
Abaluck, J. and J. Gruber (2011). Choice inconsistencies among the elderly: Evidence from plan
choice in the Medicare Part D program. American Economic Review 101(4), 1180–1210.
27
Abaluck, J. and J. Gruber (2016). Choice inconsistencies among the elderly: Evidence from plan
choice in the Medicare Part D program: Reply. American Economic Review 106(12), 3962–87.
American Heart Association (2017). High blood pressure redefined for first time in 14 years: 130 is
the new high.http://newsroom.heart.org/news/, retrieved March 5, 2018.
Andersen, H. H., A. M¨
uhlbacher, M. N ¨
ubling, J. Schupp, and G. G. Wagner (2007). Computa-
tion of standard values for physical and mental health scale scores using the SOEP version
of SF12v2. Journal of Applied Social Science Studies (Schmollers Jahrbuch) 127, 171–182.
Avery, M., O. Giuntella, and P. Jiao (2019). Why don’t we sleep enough? A field experiment
among college students. IZA Discussion Papers 12772.
Babcock, L. and G. Loewenstein (1997). Explaining bargaining impasse: The role of self-serving
biases. The Journal of Economic Perspectives 11(1), 109–126.
Bago d’Uva, T., O. O’Donnell, and E. van Doorslaer (2008). Differential health reporting by
education level and its impact on the measurement of health inequalities among older Euro-
peans. International Journal of Epidemiology 37(6), 1375–1383.
Bago d’Uva, T., O. O’Donnell, and E. van Doorslaer (2020). Who can predict their own demise?
Heterogeneity in the accuracy and value of longevity expectations. The Journal of the Eco-
nomics of Ageing forthcoming.
Baker, M., M. Stabile, and C. Deri (2004). What do self-reported, objective, measures of health
measure? Journal of Human Resources 39(4), 1067–1093.
Barber, B. M. and T. Odean (2001). Boys will be boys: Gender, overconfidence, and common
stock investment. Quarterly Journal of Economics 116(1), 261–292.
Becker, G. S., M. Grossman, and K. M. Murphy (1994). An empirical analysis of cigarette ad-
diction. American Economic Review 84(3), 396–418.
Becker, G. S. and K. M. Murphy (1988). A theory of rational addiction. Journal of Political
Economy 96(4), 675–700.
Belot, M., J. James, and J. Spiteri (2019). Facilitating healthy dietary habits: An experiment with
a low income population. IZA Discussion Papers 12675.
B´
enabou, R. and J. Tirole (2002). Self-confidence and personal motivation. The Quarterly Journal
of Economics 117(3), 871–915.
Benoˆ
ıt, J.-P. and J. Dubra (2011). Apparent overconfidence. Econometrica 79(5), 1591–1625.
Bertram, L., A. B¨
ockenhoff, I. Demuth, S. D¨
uzel, R. Eckardt, S.-C. Li, U. Lindenberger, G. Paw-
elec, T. Siedler, G. G. Wagner, and E. Steinhagen-Thiessen (2014). Cohort profile: The Berlin
Aging Study II (BASE-II). International Journal of Epidemiology 43(3), 703–712.
Bhargava, S., G. Loewenstein, and J. Sydnor (2017). Choose to lose: Health plan choices from a
menu with dominated option. The Quarterly Journal of Economics 132(3), 1319–1372.
Blanchflower, D. G., B. van Landeghem, and A. J. Oswald (2009). Imitative Obesity and Relative
Utility. Journal of the European Economic Association 7(2-3), 528–538.
B¨
ockenhoff, A., D. Saßenroth, M. Kroh, T. Siedler, P. Eibich, and G. G. Wagner (2013). The
socio-economic module of the Berlin Aging Study II (SOEP-BASE): Description, structure,
and questionnaire. SOEPpapers on Multidisciplinary Panel Data Research 568, DIW Berlin,
The German Socio-Economic Panel (SOEP).
28
Burks, S. V., J. P. Carpenter, L. Goette, and A. Rustichini (2013). Overconfidence is a social
signaling bias. Review of Economic Studies 80(3), 949–983.
Camerer, C. and D. Lovallo (1999). Overconfidence and excess entry: An experimental ap-
proach. The American Economic Review 89(1), pp. 306–318.
Carrera, M., H. Royer, M. Stehr, J. Sydnor, and D. Taubinsky (2018). The limits of simple imple-
mentation intentions: Evidence from a field experiment on making plans to exercise. Journal
of Health Economics 62(C), 95–104.
Carrera, M., H. Royer, M. F. Stehr, and J. R. Sydnor (2020). The structure of health incentives:
Evidence from a field experiment. Management Science 66(5), 1890–1908.
Cawley, J. and T. Philipson (1999). An empirical examination of information barriers to trade
in insurance. American Economic Review 89(4), 827–846.
Cawley, J. and C. Ruhm (2011). The economics of risky health behaviors. In Handbook of Health
Economics, Volume 2, Chapter Three, pp. 95–199. Elsevier.
Chaloupka, F. (1991). Rational addictive behavior and cigarette smoking. Journal of Political
Economy 99(4), 722–742.
Charness, G., J. van de Ven, and A. Rustichini (2018). Self-confidence and strategic behavior.
Experimental Economics 21(1), 72–98.
Choi, A. and J. Cawley (2018). Health disparities across education: The role of differential
reporting error. Health Economics 27(3), e1–e29.
Cowan, B. W. (2018). Sources of bias in teenagers’ college expectations. Social Science Quar-
terly 99(1), 136–153.
Darden, M. (2017). Smoking, expectations, and health: A dynamic stochastic model of lifetime
smoking behavior. Journal of Political Economy 125(5), 1465–1522.
Davillas, A. and S. Pudney (2017). Concordance of health states in couples: Analysis of self-
reported, nurse administered and blood-based biomarker data in the UK Understanding
Society panel. Journal of Health Economics 56, 87–102.
Della Vigna, S. and U. Malmendier (2006). Paying not to go to the gym. American Economic
Review 96(3), 694–719.
Di Girolamo, A., G. W. Harrison, M. I. Lau, and J. T. Swarthout (2015). Subjective belief distri-
butions and the characterization of economic literacy. Journal of Behavioral and Experimental
Economics (formerly The Journal of Socio-Economics) 59(C), 1–12.
Di Tella, R., R. Perez-Truglia, A. Babino, and M. Sigman (2015). Conveniently upset: Avoiding
altruism by distorting beliefs about others’ altruism. American Economic Review 105(11), 3416–
42.
Dohmen, T., A. Falk, D. Huffman, U. Sunde, J. Schupp, and G. G. Wagner (2011). Individual risk
attitudes: Measurement, determinants, and behavioral consequences. Journal of the European
Economic Association 9(3), 522–550.
Dragone, D. (2009). A rational eating model of binges, diets and obesity. Journal of Health
Economics 28(4), 799 – 804.
Dragone, D. and N. R. Ziebarth (2017). Non-separable time preferences, novelty consump-
tion and body weight: Theory and evidence from the East German transition to capitalism.
Journal of Health Economics 51, 41 – 65.
29
Etil´
e, F. and C. Milcent (2006). Income-related reporting heterogeneity in self-assessed health:
evidence from France. Health Economics 15(9), 965–981.
Feinleib, M. (1985). Epidemiology of obesity in relation to health hazards. Annals of Internal
Medicine 103(6 Part 2), 1019–1024.
Friehe, T. and M. Pannenberg (2019). Overconfidence over the lifespan: Evidence from Ger-
many. Journal of Economic Psychology 74, 102207.
Giuntella, O. and F. Mazzonna (2019). Sunset time and the economic effects of social jetlag:
evidence from us time zone borders. Journal of Health Economics 65, 210 – 226.
Glanz, K. (1988). Patient and public education for cholesterol reduction: a review of strategies
and issues. Patient Education and Counseling 12(3), 235 – 257.
Goette, L., S. Bendahan, J. Thoresen, F. Hollis, and C. Sandi (2015). Stress pulls us apart: trait
anxiety modulates the response of self-confidence to stress. Psychoneuroendocrinology 54, 115
– 123.
Grossman, M. (1972). On the concept of health capital and the demand for health. Journal of
Political Economy 80(2), 223–255.
Gruber, J. and B. K˝
oszegi (2001). Is addiction “rational”? Theory and evidence. Quarterly
Journal of Economics 116(4), 1261–1303.
Harris, M. C. (2017). Imperfect information on physical activity and caloric intake. Economics
& Human Biology 26(C), 112–125.
Harrison, G. and E. Rutstr¨
om (2006). Eliciting subjective beliefs about mortality risk orderings.
Environmental & Resource Economics 33(3), 325–346.
Harrison, G. W. (2015). Eliciting Subjective Beliefs of Health Risks. mimeo.
Harrison, G. W., J. Mart´
ınez-Correa, J. T. Swarthout, and E. R. Ulm (2017). Scoring rules for
subjective probability distributions. Journal of Economic Behavior & Organization 134(C), 430–
448.
Harrison, G. W., J. Mart ´
nez-Correa, J. T. Swarthout, and E. R. Ulm (2015). Eliciting subjective
probability distributions with binary lotteries. Economics Letters 127(C), 68–71.
Heidhues, P., B. K˝
oszegi, and P. Strack (2019). Overconfidence and prejudice. memo.
Himmler, O. and T. Koenig (2012). Self-evaluations and performance: Evidence from
adolescence. Hannover Economic Papers (HEP) dp-507, Leibniz Universit ¨
at Hannover,
Wirtschaftswissenschaftliche Fakult¨
at.
Holst, C., U. Becker, M. Jørgensen, M. Grønbæk, and T. J. S. (2017). Alcohol drinking patterns
and risk of diabetes: a cohort study of 70,551 men and women from the general Danish
population. Diabetologia 60(10), 1941–195.
Institute of Medicine (1990). Physical fitness and exercise. In M. Stoto, R. Behrens, and
C. Rosemont (Eds.), Healthy People 2000: Citizens Chart the Course, Chapter 13. National
Academies Press. Available from: https://www.ncbi.nlm.nih.gov/books/NBK235774/, re-
trieved November 11, 2020.
Jin, L. and N. R. Ziebarth (2020). Sleep, health, and human capital: Evidence from daylight
saving time. Journal of Economic Behavior & Organization 170, 174 – 192.
J¨
urges, H. (2007). True health vs response styles: exploring cross-country differences in self-
reported health. Health Economics 16(2), 163–178.
30
J¨
urges, H. (2008). Self-assessed health, reference levels and mortality. Applied Economics 40(5),
569–582.
Kerkhofs, M. and M. Lindeboom (1995). Subjective health measures and state dependent re-
porting errors. Health Economics 4(3), 221–235.
Ketcham, J. D., N. V. Kuminoff, and C. A. Powers (2016). Choice inconsistencies among the
elderly: Evidence from plan choice in the Medicare Part D program: Comment. American
Economic Review 106(12), 3932–61.
Ketcham, J. D., C. Lucarelli, E. J. Miravete, and M. C. Roebuck (2012). Sinking, swimming, or
learning to swim in Medicare Part D. American Economic Review 102(6), 2639–73.
Kettlewell, N. (2020). Policy choice and product bundling in a complicated health insurance
market: Do people get it right? Journal of Human Resources 55(2), 566–610.
K˝
oszegi, B. (2006). Ego utility, overconfidence, and task choice. Journal of the European Economic
Association 4(4), 673–707.
Lindeboom, M. and E. van Doorslaer (2004). Cut-point shift and index shift in self-reported
health. Journal of Health Economics 23(6), 1083–1099.
Linn´
er, R. K., P. Biroli, E. Kong, S. F. W. Meddens, R. Wedow, M. A. Fontana, M. Lebreton,
A. Abdellaoui, A. R. Hammerschlag, M. G. Nivard, A. Okbay, C. A. Rietveld, P. N. Timshel,
S. P. Tino, M. Trzaskowski, R. d. Vlaming, C. L. Z ¨
und, Y. Bao, L. Buzdugan, A. H. Caplin, C.-
Y. Chen, P. Eibich, P. Fontanillas, J. R. Gonzalez, P. K. Joshi, V. Karhunen, A. Kleinman, R. Z.
Levin, C. M. Lill, G. A. Meddens, G. Muntan´
e, S. Sanchez-Roige, F. J. v. Rooij, E. Taskesen,
Y. Wu, F. Zhang, A. Auton, J. D. Boardman, D. W. Clark, A. Conlin, C. C. Dolan, U. Fis-
chbacher, P. J. F. Groenen, K. M. Harris, G. Hasler, A. Hofman, M. A. Ikram, S. Jain, R. Karls-
son, R. C. Kessler, M. Kooyman, J. MacKillop, M. M¨
annikk¨
o, C. Morcillo-Suarez, M. B. Mc-
Queen, K. M. Schmidt, M. C. Smart, M. Sutter, A. R. Thurik, A. G. Uitterlinden, J. White,
H. d. Wit, J. Yang, L. Bertram, D. Boomsma, T. Esko, E. Fehr, D. A. Hinds, M. Johannesson,
M. Kumari, D. Laibson, P. K. E. Magnusson, M. N. Meyer, A. Navarro, A. A. Palmer, T. H.
Pers, D. Posthuma, D. Schunk, M. B. Stein, R. Svento, H. Tiemeier, P. R. H. J. Timmers, P. Tur-
ley, R. J. Ursano, G. G. Wagner, J. F. Wilson, J. Gratten, J. J. Lee, D. Cesarini, D. J. Benjamin,
P. D. Koellinger, and J. P. Beauchamp (2019). Genome-wide association analyses of risk toler-
ance and risky behaviors in over 1 million individuals identify hundreds of loci and shared
genetic influences. Nature Genetics 51, 245–257.
Lundborg, P. (2007). Smoking, information sources, and risk perceptions—new results on
Swedish data. Journal of Risk and Uncertainty 34(3), 217–240.
Maguire, C. P., J. Ryan, A. Kelly, D. O’Neill, D. Coakley, and J. B. Walsh (2000). Do patient
age and medical condition influence medical advice to stop smoking? Age and Ageing 29(3),
264–266.
Mathieu-Bolh, N. and R. Wendner (2020). We are what we eat: Obesity, income, and social
comparisons. European Economic Review 128, 103495.
Moore, D. A. and P. J. Healy (2008). The trouble with overconfidence. Psychological review 115(2),
502.
National Health Service (2012). Have two alcohol-free days a week, say MPs.hhttps://www.
nhs.uk/news/lifestyle-and-exercise/avoid-alcohol- 2-days-per- week/,
retrieved June 15, 2020.
31
National Institutes of Health (2000). The Practical Guide to the Identification, Evaluation, and Treat-
ment of Overweight and Obesity in Adults.https://www.nhlbi.nih.gov/files/docs/
guidelines/prctgd_c.pdf, retrieved June 15, 2020.
Ortoleva, P. and E. Snowberg (2015). Overconfidence in political behavior. American Economic
Review 105(2), 504–35.
Rabin, M. (2013, June). Incorporating limited rationality into economics. Journal of Economic
Literature 51(2), 528–43.
RAND (1995). User’s manual for the medical outcomes study (MOS) core measures of health-
related quality of life. Technical report.
Rehm, J., T. K. Greenfield, and J. D. Rogers (2001). Average volume of alcohol consumption,
patterns of drinking, and all-cause mortality: Results from the us national alcohol survey.
American Journal of Epidemiology 153(1), 64–71.
Renaud, S. C., R. Gu´
eguen, G. Siest, and R. Salamon (1999). Wine, beer, and mortality in middle-
aged men from eastern France. Archives of Internal Medicine 159(16), 1865–1870.
Richter, D., M. Metzing, M. Weinhardt, and J. Schupp (2013). SOEP Scales Manual—
Erhebungsinstrumente Berliner Altersstudie II. SOEP Survey Papers: Series C -
Data Documentations (Datendokumentationen) 138. panel.gsoep.de/soep-docs/
surveypapers/diw_ssp0138.pdf, retrieved January 17, 2014.
Richter, D. and J. Schupp (2015). The SOEP Innovation Sample (SOEP IS). Schmollers
Jahrbuch 135(3), 389–399.
Richter, D. and J. Schupp (2017). Questionnaire for the SOEP Innovation sample (boost sample).
SOEP Survey Papers 454, DIW Berlin, The German Socio-Economic Panel (SOEP).
Robert Koch Institut (1995). Gesundheitssurvey Ost/West, Befragungs- und Untersuchungssurvey
in den neuen und alten Bundesl¨andern. Robert Koch Institut. ublic Use File OW91 (1990 - 1992),
Dokumentation des Datensatzes zusammengestellt von Dr. Heribert Stolzenberg.
Robert Koch Institut (2012). German National Health Survey East-West 1991 (GNHSEW91),
Public Use File OW91. Robert Koch Institut. http://www.rki.de/DE/Content/
Gesundheitsmonitoring/PublicUseFiles/informationen_datensaetze/
info_datensaetze_node.html, last accessed on June 11, 2016.
Royer, H., M. Stehr, and J. Sydnor (2015). Incentives, commitments, and habit formation in
exercise: Evidence from a field experiment with workers at a fortune-500 company. American
Economic Journal: Applied Economics 7(3), 51–84.
Sandroni, A. and F. Squintani (2007). Overconfidence, insurance, and paternalism. The American
Economic Review 97(5), 1994–2004.
Santos-Pinto, L. and J. Sobel (2005). A model of positive self-image in subjective assessments.
American Economic Review 95(5), 1386–1402.
Sharot, T. (2011). The optimism bias. Current biology 21(23), R941–R945.
Spinnewijn, J. (2015). Unemployed but optimistic: Optimal insurance design with biased be-
liefs. Journal of the European Economic Association 13(1), 130–167.
Spitzer, S. and D. Weber (2019). Who is telling the truth? Biases in self-reported physical and
cognitive health status of older Europeans. PLoS ONE 10(14), e0223526.
32
Strulik, H. and T. Trimborn (2018). Hyperbolic discounting can be good for your health. Cen-
ter for European, Governance and Economic Development Research Discussion Papers 335,
University of Goettingen, Department of Economics.
Tiefenbeck, V., L. Goette, K. Degen, R. Lalive, and T. Staake (2018). Overcoming salience bias:
How real-time feedback fosters resource conservation. Management Science 64(3), 1458–1476.
Trichopoulos, D., A. Kalandidi, L. Sparros, and B. Macmahon (1981). Lung cancer and passive
smoking. International Journal of Cancer 27(1), 1–4.
Viscusi, W. K. (1990). Do smokers underestimate risks? Journal of Political Economy 98(6), 1253–
1269.
Wagner, G. G., J. R. Frick, and J. Schupp (2007). The German Socio-Economic Panel Study
(SOEP)–scope, evolution and enhancements. Schmollers Jahrbuch : Journal of Applied Social
Science Studies / Zeitschrift f¨ur Wirtschafts- und Sozialwissenschaften 127(1), 139–169.
Weinberg, B. A. (2006). A model of overconfidence. Working paper, Ohio State University.
Weinstein, N. (1989). Optimistic biases about personal risks. Science 246(4935), 1232–1233.
Weinstein, N. D. (1980). Unrealistic optimism about future life events. Journal of Personality and
Social Psychology 39(5), 806.
Weiss, N. S. (1972). Relation of high blood pressure to headache, epistaxis, and selected other
symptoms. New England Journal of Medicine 287(13), 631–633.
Ziebarth, N. R. (2010). Measurement of health, health inequality, and reporting heterogeneity.
Social Science & Medicine 71, 116–124.
Ziebarth, N. R. (2018). Lung cancer risk perception biases. Preventive Medicine 110, 16–23.
Ziebarth, N. R. and M. M. Grabka (2009). In vino pecunia? The association between beverage-
specific drinking behavior and wages. Journal of Labor Research 30(3), 219–240.
Ziebarth, N. R. and M. Karlsson (2010). A natural experiment on sick pay cuts, sickness absence,
and labor costs. Journal of Public Economics 94(11–12), 1108–1122.
Zwald, M. L., B. K. Kit, T. H. Fakhouri, J. P. Hughes, and L. J. Akinbami (2019). Prevalence
and correlates of receiving medical advice to increase physical activity in U.S. adults: Na-
tional Health and Nutrition Examination Survey 2013–2016. American Journal of Preventive
Medicine 56(6), 834 – 843.
33
Figures
Figure 1: Absolute Health Perception Bias about High Cholesterol Levels
(a) High Cholesterol Levels (b) High Blood Pressure
6.5% 14.2%
29.8%
49.5%
0 1
Hi
0 1
Hi
~
13.5% 11.6%
9.3%
65.5%
0 1
Hi
0 1
Hi
~
Source: GNHSEW91. See Section 4.3 and 5for more details. The total number of observations is 6,429.
The number of observations in Figure 1a, counting clockwise and starting in the upper left corner are
420, 912, 3184, and 1913. In Figure 1b, the number of observations are 873, 746, 4209, and 601.
Figure 2: Perceived Population Share in Better Health ( e
bi)
0 .05 .1 .15 .2 .25
Fraction
0 10 20 30 40 50 60 70 80 90 100
Number of respondents with better health status
0 .05 .1 .15 .2 .25
Fraction
0 10 20 30 40 50 60 70 80 90 100
Number of respondents with better health status
Sources: BASE-II (left panel), SOEP-IP (right panel). Responses to the question are plotted: “Imagine one would
randomly select 100 people in your age. How many of those 100 people would be in better health than you?” People
answering 0 believe nobody is in better health; people answering 99 believe everybody is healthier than them.”
34
Figure 3: Actual (ri) and Perceived ( ˜
ri) Ranking in Population Health Distribution
BASE-II SOEP-IP
0 10 20 30 40 50 60 70 80 90 100
ri
0 10 20 30 40 50 60 70 80 90 100
ri
~
Ri<0
Ri>0
0 10 20 30 40 50 60 70 80 90 100
ri
0 10 20 30 40 50 60 70 80 90 100
ri
~
Ri<0
Ri>0
Source: BASE-II, SOEP-IP. Relative health perception biases are defined as Ri=e
riri, with ˜
ri= 1 bi,
see Figure 2and main text. The true rank in the population health distribution, ri, is based on the SF12.
The x-axes indicate e
riand the y-axes indicate riwhere the true population health ranks riare averaged
by e
ri-bins of ten ranks. The size of the scatters indicate the number of respondents in each bin. For
example, in Figure 3a, we observe a mass of respondents who believe that they rank between the 70th
and 90th percentile of the population health distribution but whose actual health status is just average.
Figure 4: Distribution of RiBased on (a) SAH, (b) SF12 in SOEP-IP
0 .005 .01 .015 .02
Density
-100 100-50 0 50
Health perception bias SAH (rank differences)
males
females
kernel = epanechnikov, bandwidth = 9.8000
0 .005 .01 .015
Density
-100 100
-50 0 50
Health perception bias SF12 (rank differences)
males
females
kernel = epanechnikov, bandwidth = 9.8000
Source: SOEP-IP. Figure displays distributions of Ri=e
ririin the representative SOEP-IP, with ˜
ri= 1 e
bi, see
Figure 2and main text. Subfigure (a) uses SAH and subfigure (b) uses the SF12 as ri.
35
Figure 5: Absolute Health Perception Bias (High Cholesterol) and Risky Health Behavior
0 .2 .4 .6
No Sport
26 27 28
BMI
0 .2 .4 .6
Alcohol Daily
A<=0 A>0
0 .2 .4 .6
Smoker
A<=0 A>0
Source: GNHSEW91. Bar diagrams show, along with 95% confidence intervals, the (a) share of respondents who
do not exercise, (b) mean BMI, as well as the share of respondents who (c) drink alcohol daily and (d) smoke. The
light gray bars and ’1’ indicate respondents who exhibit absolute health perception biases (N=3184) and the dark
gray bars and ’0’ indicate respondents who do not exhibit absolute health perception biases with respect to high
blood cholesterol levels (N=4096).
36
Figure 6: Absolute Health Perception Bias (High Blood Pressure) and Risky Health Behavior
0 .2 .4 .6
No Sport
26 27 28
BMI
0 .2 .4 .6
Alcohol Daily
A<=0 A>0
0 .2 .4 .6
Smoker
A<=0 A>0
Source: GNHSEW91. Bar diagrams show, along with 95% confidence intervals, the (a) share of respondents who
do not exercise, (b) mean BMI, as well as the share of respondents who (c) drink alcohol daily and (d) smoke. The
light gray bar and ’1’ indicate respondents who exhibit absolute health perception biases (N=601) and the dark gray
bar and ’0’ indicate respondents who do not exhibit absolute health perception biases with respect to high blood
pressure levels (N=4955).
37
Figure 7: Relative Health Perception Bias and Risky Health Behavior
.2 .3 .4 .5 .6
No sports
-25-20 -15 -10 -5 0 5 10 15 20 25 30 35 40 45 50
health perception bias
(deviation from true population rank)
95% CI lpoly smooth
kernel = epanechnikov, degree = 0, bandwidth = 6.98, pwidth = 10.47
25 25.5 26 26.5 27
BMI
-25-20 -15 -10 -5 0 5 10 15 20 25 30 35 40 45 50
health perception bias
(deviation from true population rank)
95% CI lpoly smooth
kernel = epanechnikov, degree = 0, bandwidth = 5.38, pwidth = 8.07
.3 .35 .4 .45 .5
No healthy diet
-25-20 -15 -10 -5 0 5 10 15 20 25 30 35 40 45 50
health perception bias
(deviation from true population rank)
95% CI lpoly smooth
kernel = epanechnikov, degree = 0, bandwidth = 8.92, pwidth = 13.38
.05 .1 .15 .2
Smoker
-25-20 -15 -10 -5 0 5 10 15 20 25 30 35 40 45 50
health perception bias
(deviation from true population rank)
95% CI lpoly smooth
kernel = epanechnikov, degree = 0, bandwidth = 6.11, pwidth = 9.16
Source: BASE-II. Figure shows non-parametric kernel-weighted local polynomial smoothing plots. The y-axis
shows (a) the likelihood that respondents do not exercise, (b) are obese (BMI>30), (c) follow an unhealthy diet,
or (d) smoke. The x-axis indicates Ri, that is, the deviation of the perceived rank (e
ri) in the population health
distribution from the true rank (ri).
38
Figure 8: Relative Health Perception Biases and the Sleep Gap to 8 Hours (SOEP-IP)
.6 .8 1 1.2 1.4 1.6
sleep gap week
-25-20 -15 -10 -5 0 5 10 15 20 25 30 35 40 45 50
health perception bias
(deviation from true population rank)
95% CI lpoly smooth
kernel = epanechnikov, degree = 0, bandwidth = 4.24, pwidth = 6.36
0 .2 .4 .6 .8
sleep gap weekend
-25-20 -15 -10 -5 0 5 10 15 20 25 30 35 40 45 50
health perception bias
(deviation from true population rank)
95% CI lpoly smooth
kernel = epanechnikov, degree = 0, bandwidth = 5.77, pwidth = 8.65
Source: SOEP-IP. Figure shows non-parametric kernel-weighted local polynomial smoothing plots. The y-axis
shows the difference between 8 hours of sleep and actual hours of sleep (a) during the week, (b) on weekends. The
x-axis indicates Ri, that is, the deviation of the perceived rank (e
ri) in the population health distribution from the
true rank (ri).
39
Table 1: Relative Perception Bias and Risky Health Behaviors: No Sports, Obesity, No Healthy Diet, Smoking
(1) (2) (3) (4) (5) (6) (7) (8)
No sports Obese No healthy diet Smoker
Positive health bias (Ri>0)0.0023*** 0.0022*** 0.0011*** 0.0012*** 0.0013** 0.0012* 0.0005 0.0004
(0.0006) (0.0006) (0.0004) (0.0004) (0.0006) (0.0006) (0.0004) (0.0004)
Negative health bias (Ri<0)-0.0003 -0.0008 -0.0010 -0.0011 -0.0006 -0.0007 0.0003 0.0002
(0.0014) (0.0015) (0.0010) (0.0010) (0.0014) (0.0015) (0.0009) (0.0009)
R20.0276 0.0380 0.0298 0.0340 0.0492 0.0579 0.0884 0.0925
socio-demographics & education yes yes yes yes yes yes yes yes
employment char. & income no yes no yes no yes no yes
month FE yes yes yes yes yes yes yes yes
Source: Berlin Aging Study II (BASE-II); * p<0.1, ** p<0.05, *** p<0.01; standard errors in parentheses. The descriptive statistics
are in the Appendix (Table A2). The model is estimated by OLS with n=1,804 observations. The binary dependent variables in
columns (1) to (8) measure the likelihood that a respondent does not exercise at all, that a respondent is obese (BMI>30), that a
respondent follows an unhealthy diet and that a respondent is a current smoker. Positive health bias (Ri>0) and negative health
bias (Ri<0) are continuous health bias measures, using SF12 to measure Hi. For more information, see Sections 4.2 and 5.
Table 2: Relative Perception Bias and Sleep Gap to 8 Hours
(1) (2) (3) (4)
Sleep Gap to 8 Hours (SF12) (SF12) (SAH) (SAH)
Positive health bias (Ri>0)0.0037* 0.0038* 0.0068*** 0.0066***
(0.0021) (0.0021) (0.0025) (0.0025)
Negative health bias (Ri<0)0.0089** 0.0086** 0.0028 0.0028
(0.0037) (0.0038) (0.0025) (0.0025)
R20.0455 0.0513 0.0417 0.0475
sociodem. & educ. yes yes yes yes
employment char. & income no yes no yes
month FE yes yes yes yes
Source: SOEP-IP; * p<0.1, ** p<0.05, *** p<0.01; standard errors in parentheses.
The descriptive statistics are in the Appendix (Table A3). The model is estimated by
OLS with n=1,397 observations; the dependent variable measures the gap between
the actual hours of sleep during the week and eight hours. Positive health bias
(Ri>0) and negative health bias (Ri<0) are continuous health bias measures.
More information on the variables, see Sections 4.3 and 5.
41
Table 3: Robustness Check: Additional Covariates: Risk Aversion and Personality Traits
(1) (2) (3) (4) (5) (6) (7) (8)
No sports Obese No healthy diet Smoker
Positive health bias (Ri>0)0.0021*** 0.0018*** 0.0012*** 0.0012*** 0.0011* 0.0009 0.0004 0.0005
(0.0006) (0.0006) (0.0004) (0.0004) (0.0006) (0.0006) (0.0004) (0.0004)
Negative health bias (Ri<0)-0.0008 -0.0009 -0.0011 -0.0011 -0.0007 -0.0013 0.0002 0.0002
(0.0015) (0.0015) (0.0010) (0.0010) (0.0014) (0.0014) (0.0009) (0.0009)
R20.0441 0.0481 0.0351 0.0424 0.0669 0.1012 0.0947 0.0981
socio-demographics & education yes yes yes yes yes yes yes yes
employment char. & income yes yes yes yes yes yes yes yes
month FE yes yes yes yes yes yes yes yes
risk aversion indicators yes yes yes yes yes yes yes yes
personality traits no yes no yes no yes no yes
Source: Berlin Aging Study II (BASE-II); * p<0.1, ** p<0.05, *** p<0.01; standard errors in parentheses. The descriptive statistics are in the
Appendix (Table A2). This robustness check adds, for each health behavior outcome, risk aversion indicators (risk-averse is 3 and risk-loving
is 8 on a risk-aversion scale ranging from 0 to 10) and the Big 5 personality traits as additional covariates. The model is estimated by OLS with
n=1,804 observations; the columns indicate the dependent variables. Positive health bias (Ri>0) and negative health bias (Ri<0) are continuous
health bias measures, using SF12 to measure Hi. More information on the variables, see Sections 4.3 and 5.
42
Table 4: Robustness Check: Trimmed RiDistribution and Omitting Potential Guesses
(1) (2) (3) (4) (5) (6) (7) (8)
No sports Obese No healthy diet Smoker
Positive health bias (Ri>0)0.0020*** 0.0020*** 0.0008* 0.0009** 0.0013** 0.0011* 0.0005 0.0004
(0.0006) (0.0006) (0.0004) (0.0004) (0.0006) (0.0006) (0.0004) (0.0004)
Negative health bias (Ri<0)-0.0011 -0.0017 -0.0013 -0.0014 -0.0013 -0.0014 -0.0002 -0.0003
(0.0017) (0.0017) (0.0012) (0.0012) (0.0017) (0.0017) (0.0012) (0.0012)
R20.0213 0.0319 0.0253 0.0299 0.0469 0.0556 0.0726 0.0768
socio-demographics & education yes yes yes yes yes yes yes yes
employment char. & income no yes no yes no yes no yes
month FE yes yes yes yes yes yes yes yes
Source: Berlin Aging Study II (BASE-II); * p<0.1, ** p<0.05, *** p<0.01; standard errors in parentheses. The descriptive statistics
are in the Appendix (Table A2). This robustness check trims the Ridistribution and eliminates the top and bottom quintiles; we
also disregard respondents with e
ri=50, who presumably guessed and may have low confidence in their estimate (the results are
robust to not eliminating these respondents). The model is estimated by OLS with n=1,539 observations; the columns indicate the
dependent variables. Positive health bias (Ri>0) and negative health bias (Ri<0) are continuous health bias measures, using
SF12 to measure Hi. More information on the variables, see Section 4.2 and 5.
43
Appendix
Figure A1: Absolute Health Perception Bias Aiin East vs. West Germany
0 .2 .4 .6
East Germany
West Germany
0 .2 .4 .6
East Germany
West Germany
Source: GNHSEW91. Bar diagrams show, along with 95% confidence intervals, absolute health perception biases for East
and West Germans for (a) blood cholesterol levels, (b) high blood pressure. The light gray bars indicate East German
respondents (N= 2,125) and the dark gray bars indicate West German respondents (N=4,304).
44
Figure A2: Distribution of SAH and SF12 in BASE-II
0 .1 .2 .3 .4 .5
Fraction
0 1 2 3 4 5
Subjectively assessed health status (SAH)
0 .02 .04 .06
Density
20 30 40 50 60
Distribution of SF12
Figure A3: Distribution of SAH and SF12 in SOEP-IP
0 .1 .2 .3 .4
Density
0 1 2 3 4 5
sah
0 .02 .04 .06 .08
Density
20 30 40 50 60
SF12
45
Figure A4: Distribution of RiBased on (a) SAH, (b) SF12 in BASE-II
0 .005 .01 .015 .02
Density
-100 -50 0 50 100
Health perception bias SAH (rank differences)
males
females
kernel = epanechnikov, bandwidth = 9.8000
0 .005 .01 .015
Density
-100 -50 0 50 100
Health perception bias SF12 (rank differences)
males
females
kernel = epanechnikov, bandwidth = 9.8000
Source: BASE-II. Figure displays distributions of Ri=e
riri, with ˜
ri= 1 e
bi, see Figure 2and main text. Subfigure (a)
uses SAH and subfigure (b) uses the SF12 as ri.
46
Table A1: Descriptive Statistics German National Health Survey East-West 1991
Variable Mean Std. Dev. Min. Max. N
A. Health Bias Measures
Absolute Health Bias Cholesterol, Ai>1 0.2976 0.4572 0 1 6429
Absolute Health Bias Blood Pressure, Ai>1 0.0935 0.2911 0 1 6429
B. Objective Health Measures (Hi)
Total blood cholesterol [mmol/l] 6.13 1.23 2.33 12.9 6429
High total blood cholesterol [>6.2 mmol/l] 0.4394 0.4964 0 1 6429
Systole, 2. measure [mmHg] 134.66 20.263 88 256 6429
Diastole, 2. measure [mmHg] 83.40 12.129 34 158 6429
Hypertension 0.2095 0.407 0 1 6429
C. Subjective Health Assessment ( ˜
Hi)
High Cholesterol 0.2072 0.4053 0 1 6429
High Blood Pressure 0.2518 0.4341 0 1 6429
D. Health Behavior
Alcohol Daily 0.1618 0.3683 0 1 6429
Current Smoker 0.3282 0.4696 0 1 6429
Body-mass-index [kg per m2] 26.65 4.6113 15.02 75.47 6429
Obese (BMI>30) 0.2019 0.4014 0 1 6429
No sports 0.4652 0.4988 0 1 6429
Sources: GNHSEW91, own illustration.[mmol/l] stands for millimole per liter. [mmHg] stands for
millimetres of mercury. [kg per m2] stands for kilogram per square meter.
47
Table A2: Descriptive Statistics BASE-II
Variable Mean Std. Dev. Min. Max. N
A. Health Bias Measures
Relative Health Bias SF12, Ri19.83 25.95 -66.06 93.65 1804
Positive (Ri>0), SF12 23.03 21.36 0 93.65 1804
Negative (Ri<0), SF12 -3.19 8.37 -66.06 0 1804
Relative Health Bias SAH, Ri3.7605 26.24 -98 95.77 1804
Positive (Ri>0), SAH 11.47 17.23 0 95.12 1804
Negative (Ri<0), SAH -2.1508 9.7383 -98 0 1804
B. Health Behavior
No sports 0.3564 0.4791 0 1 1804
Obese 0.1264 0.3324 0 1 1804
BMI 25.51 4.3106 13.71 64.09 1804
No healthy diet 0.3858 0.4869 0 1 1804
Smoker 0.1175 0.3221 0 1 1804
C. Covariates
Demographics
Age 60.35 16.81 18.18 89.98 1804
Female 0.5244 0.4995 0 1 1804
Married 0.5676 0.4955 0 1 1804
Single 0.2506 0.4335 0 1 1804
Partner in Household 0.6547 0.4756 0 1 1804
# kids 1.3126 1.1357 0 5 1804
# daughters 0.6613 0.8307 0 4 1804
No kids 0.3099 0.4626 0 1 1804
German 0.9878 0.1098 0 1 1804
Education
8 school years 0.1292 0.3355 0 1 1804
10 school years 0.2517 0.4341 0 1 1804
13 school years 0.5133 0.5 0 1 1804
Employment & Income
Blue collar worker 0.0272 0.1626 0 1 1804
White collar worker 0.189 0.3916 0 1 1804
Civil servant 0.0183 0.134 0 1 1804
Full-time employed 0.4678 0.4991 0 1 1804
Part-time employed 0.1414 0.3485 0 1 1804
Gross labor earnings 549 1298 0 20,000 1804
Net labor earnings (last month) 378 831 0 10,000 1804
Total income (last month) 1563 1294 0 20950 1804
Risk Aversion
Risk aversion (scale) 5.0837 2.2255 0 10 1804
Risk-averse 0.2611 0.4393 0 1 1804
Risk-loving 0.153 0.3601 0 1 1804
Big Five
Openness 4.9916 1.15 1.3333 7 1804
Conscientiousness 5.6159 0.9726 1.6667 7 1804
Extraversion 4.7431 1.1737 1 7 1804
Neuroticism 3.7714 1.2776 1 7 1804
Agreeableness 5.2348 0.9828 1.3333 7 1804
Sources: Berlin Aging Study II (BASE-II).
48
Table A3: Descriptive Statistics SOEP-IP
Variable Mean Std. Dev. Min. Max. N
A. Health Bias Measures
Relative Health Bias SF12, Ri13.89 27.41 -97.14 98.69 1397
Ri>0, SF12 18.92 20.67 0 98.699 1397
Ri<0, SF12 -5.02 11.59 -97.14 0 1397
Relative Health Bias SAH, Ri1.0154 27.22 -100 94.49 1397
Ri>0, SAH 9.85 16.46 0 94.49 1397
Ri<0, SAH -5.46 15.15 -100 0 1397
B. Health Behavior
Sleep in hours, weekday 6.8210 1.3153 2 13 1397
Sleep deficit, week 1.179 1.3153 -5 6 1397
Sleep in hours, weekend 7.5719 1.5634 2 14 1397
Sleep deficit, weekend 0.4281 1.5634 -6 6 1397
C. Covariates
Demographics
Age 51.03 18.44 16.35 93.23 1397
Female 0.5225 0.4997 0 1 1397
Married 0.5254 0.4995 0 1 1397
Single 0.2584 0.4379 0 1 1397
# kids 0.5841 0.9445 0 5 1397
German 0.9399 0.2378 0 1 1397
Education
No degree 0.6535 0.476 0 1 1397
Apprenticeship degree 0.2004 0.4005 0 1 1397
College degree 0.1274 0.3336 0 1 1397
Employment & Income
Blue collar worker 0.073 0.2603 0 1 1397
White collar worker 0.3472 0.4762 0 1 1397
Civil servant 0.0322 0.1766 0 1 1397
Full-time employed 0.3572 0.4793 0 1 1397
Part-time employed 0.1045 0.306 0 1 1397
Gross labor earnings 1292 1833. 0 12,540 1397
Net labor earnings (last month) 878 1165 0 8000 1397
Total income (last month) 1768 1734 0 14,200 1397
Behavioral Attitudes
Risk averse (0-4/10) 0.3092 0.4623 0 1 1397
Risk loving (9-10/10) 0.1482 0.3554 0 1 1397
Risk Averse Health (0-4/10) 0.554 0.4972 0 1 1397
Risk Loving Health (9-10/10) 0.0759 0.2649 0 1 1397
Risk Averse Trust (0-4/10) 0.4409 0.4967 0 1 1397
Risk Loving Trust (9-10/10) 0.0816 0.2739 0 1 1397
Sources: SOEP-IP.
49
Table A4: Determinants of Positive and Negative Health Perception Biases (SOEP-IP)
Positive Health Perception Bias Negative Health Perception Bias
(1) (2) (3) (4)
Demographics
Age 0.035 (0.216) -0.034 (0.221) 0.125 (0.106) 0.061 (0.109)
Age2 -0.001 (0.002) 0.000 (0.002) -0.001 (0.001) -0.000 (0.001)
Female 4.329*** (1.098) 4.765*** (1.257) 2.542*** (0.657) 2.699*** (0.787)
Married -2.297 (1.506) -2.170 (1.529) 0.915 (0.900) 0.830 (0.920)
Single -3.098 (2.030) -2.912 (2.040) 1.421 (1.203) 1.480 (1.197)
# kids 1.615** (0.772) 1.488* (0.792) 0.592 (0.382) 0.697* (0.392)
German 3.665 (2.263) 3.572 (2.305) 1.863 (1.559) 1.335 (1.571)
Education
No degree 1.958 (2.042) 1.907 (2.058) -0.205 (1.056) -0.002 (1.038)
Apprenticeship degree 2.236 (2.029) 1.795 (2.070) 1.636 (1.013) 1.143 (1.002)
College degree 2.511 (2.480) 2.726 (2.511) -0.832 (1.615) -0.319 (1.574)
Employment & Inc.
Blue collar worker 1.086 (2.966) -0.013 (1.491)
White collar worker 0.082 (1.855) 1.792** (0.913)
Civil servant -1.050 (3.767) 0.142 (1.659)
Full-time employed -1.389 (2.594) -1.298 (1.214)
Part-time employed 0.751 (2.588) -2.010* (1.195)
Gross earnings -0.000 (0.001) -0.000 (0.001)
Net earnings (last mt)) 0.001 (0.002) 0.001 (0.001)
Total income (last mt) 0.000 (0.001) 0.000 (0.001)
Risk aversion
Risk averse 0.108 (1.309) -0.578 (0.679)
Risk loving -1.156 (1.674) -2.873*** (1.079)
Risk loving health 1.060 (2.329) -2.015 (1.603)
Risk averse health -1.126 (1.229) -0.651 (0.646)
Risk loving Trust 1.482 (2.292) 0.474 (1.052)
Risk averse Trust 1.836 (1.210) -0.935 (0.653)
R20.023 0.029 0.021 0.041
Source: SOEP Innovation Panel (SOEP-IP); * p<0.1, ** p<0.05, *** p<0.01; standard errors in parentheses. The
descriptive statistics are in Table A3. The model is estimated by OLS and has 1,397 observations. All models
include interview months fixed effects. The dependent variable in the first two columns use positive relative
health perception biases (Ri>0, SF12) and the second two columns use negative relative health perception
biases (Ri<0, SF12). For more information on variable generation, see Section 4.3
50
Table A5: Determinants of Positive and Negative Health Perception Biases (BASE-II)
Positive Health Perception Bias Negative Health Perception Bias
(1) (2) (3) (4)
Demographics
Age -0.840** (0.335) -1.004*** (0.312) -0.123 (0.115) -0.125 (0.113)
Age2 0.008** (0.003) 0.011*** (0.003) 0.001 (0.001) 0.002* (0.001)
Female 4.271*** (1.031) 1.597 (1.048) 0.332 (0.409) 0.051 (0.455)
Married -1.897 (1.744) -1.256 (1.657) -1.203* (0.646) -1.025 (0.643)
Single 2.547 (1.948) 2.981 (1.822) -1.007 (0.688) -0.962 (0.684)
Partner in Household 2.838 (2.135) 1.711 (2.025) -0.007 (0.766) -0.049 (0.765)
# kids 1.608* (0.825) 1.394* (0.780) 0.279 (0.318) 0.182 (0.306)
# daugthers -1.170 (0.809) -0.741 (0.752) -0.444 (0.309) -0.295 (0.297)
No kids 1.418 (1.876) 0.653 (1.774) -0.753 (0.751) -1.020 (0.745)
German 1.676 (4.658) 3.679 (4.689) 1.772 (1.976) 2.563 (2.064)
Education
8 school years -0.958 (2.159) -1.986 (1.974) -0.786 (0.729) -1.063 (0.718)
10 school years -3.772** (1.891) -3.508** (1.745) -1.649** (0.681) -1.603** (0.662)
13 school years -6.162*** (1.743) -5.425*** (1.597) -1.036* (0.578) -0.846 (0.581)
Employment & Inc.
Blue collar worker 6.778** (2.852) 0.293 (1.323)
White collar worker -0.809 (1.800) -0.568 (0.659)
Civil servant 5.445 (5.068) -1.467 (1.960)
Full-time employed -1.122 (1.527) -1.508*** (0.506)
Part-time employed -0.797 (1.863) -1.719** (0.724)
Gross earnings 0.002 (0.002) 0.002** (0.001)
Net earnings (last mt) 0.002 (0.003) -0.001 (0.001)
Total income (last mt) -0.002*** (0.001) -0.001*** (0.000)
Risk Aversion
Risk-averse -1.722 (1.113) -0.543 (0.492)
Risk-loving -1.256 (1.422) -0.472 (0.621)
Personality Traits
Openness 1.319*** (0.467) 0.239 (0.184)
Conscientiousness -0.909* (0.497) -0.639*** (0.190)
Extraversion -0.918** (0.449) -0.454** (0.195)
Neuroticism 5.456*** (0.398) 0.651*** (0.171)
Agreeableness -0.883* (0.501) -0.093 (0.212)
R20.032 0.167 0.014 0.051
Source: Berlin Aging Study II (BASE-II); * p<0.1, ** p<0.05, *** p<0.01; standard errors in parentheses. The
descriptive statistics are in the Appendix (Table A2). The model is estimated by OLS with n=1,804 observations.
All models include interview months fixed effects. Positive health bias (Ri>0) and negative health bias
(Ri<0) are continuous health bias measures, using SF12 to measure Hi. More information on the variables,
see Sections 4.2 and 5.
51
Table A6: Relative Perception Bias and Risky Health Behaviors: Using SAH to Measure Hi
(1) (2) (3) (4) (5) (6) (7) (8)
No sports Obese No healthy diet Smoker
Positive health bias (Ri>0)0.0031*** 0.0029*** 0.0022*** 0.0023*** 0.0010 0.0009 0.0005 0.0004
(0.0007) (0.0007) (0.0005) (0.0005) (0.0007) (0.0007) (0.0004) (0.0004)
Negative health bias (Ri<0)-0.0009 -0.0011 0.0001 0.0000 -0.0001 -0.0003 -0.0010 -0.0010
(0.0012) (0.0012) (0.0008) (0.0008) (0.0012) (0.0012) (0.0008) (0.0009)
R20.0301 0.0401 0.0380 0.0422 0.0476 0.0568 0.0886 0.0930
socio-demographics & education yes yes yes yes yes yes yes yes
employment char. & income no yes no yes no yes no yes
month FE yes yes yes yes yes yes yes yes
Source: Berlin Aging Study II (BASE-II); * p<0.1, ** p<0.05, *** p<0.01; standard errors in parentheses. The descriptive statistics
are in the Appendix (Table A2). The model is estimated by OLS with n=1,804 observations. The binary dependent variables in
columns (1) to (8) measure the likelihood that a respondent does not exercise at all, that a respondent is obese (BMI>30), that a
respondent follows an unhealthy diet and that a respondent is a current smoker. Positive health bias (Ri>0) and negative health
bias (Ri<0) are continuous health bias measures, using SAH to measure Hi. For more information, see Sections 4.2 and 5.
52
Table A7: Relative Perception Bias and Risky Health Behaviors: Main Results Estimated by Probit (Marginal Effects)
(1) (2) (3) (4) (5) (6) (7) (8)
No sports Obese No healthy diet Smoker
Positive health bias (Ri>0)0.0022*** 0.0021*** 0.0011*** 0.0012*** 0.0013** 0.0012** 0.0004 0.0004
(0.0006) (0.0006) (0.0004) (0.0004) (0.0006) (0.0006) (0.0004) (0.0004)
Negative health bias (Ri<0)-0.0002 -0.0007 -0.0011 -0.0011 -0.0006 -0.0007 0.0004 0.0003
(0.0015) (0.0015) (0.0010) (0.0010) (0.0015) (0.0015) (0.0009) (0.0009)
socio-demographics & education yes yes yes yes yes yes yes yes
employment char. & income no yes no yes no yes no yes
month FE yes yes yes yes yes yes yes yes
Source: Berlin Aging Study II (BASE-II); * p<0.1, ** p<0.05, *** p<0.01; standard errors in parentheses. The descriptive statistics are in the
Appendix (Table A2). The model is estimated by Probit with n=1,804 observations. The binary dependent variables in columns (1) to (8) measure
the likelihood that a respondent does not exercise at all, that a respondent is obese (BMI>30), that a respondent follows an unhealthy diet and that
a respondent is a current smoker. Positive health bias (Ri>0) and negative health bias (Ri<0) are continuous health bias measures, using SF12
to measure Hi. For more information, see Sections 4.2 and 5.
53
Table A8: Estimating the Breakpoint Rfor the Spline in the Baseline Specification
(1) (2) (3) (4) (5)
Dependent variable: No Sports Obesity Unhealthy Diet Smoker Sleep Gap
Positive health bias (Ri>0)0.0021 0.0011* 0.0013 0.0004 0.0042
(0.0016) (0.0006) (0.0010) (0.0008) (0.0049)
Negative health bias (Ri<0)-0.0043 -0.0011 -0.0011 0.0006 0.0130*
(0.0039) (0.0015) (0.0024) (0.0015) (0.0072)
socio-demographics & education yes yes yes yes yes
employment char. & income no no no no no
month FE yes yes yes yes yes
Optimal R -20.0 -1.6 -7.1 9.6 -18.7
(21.73) (13.46) (16.66) (18.57) (22.09)
p-value for likelihood ratio test 0.192 0.800 0.712 0.877 0.337
Source: Berlin Aging Study II (BASE-II); * p<0.1, ** p<0.05, *** p<0.01; standard errors in parentheses.
The descriptive statistics are in the Appendix (Table A2). The models have 1,868 observations and
estimate the breakpoint Rfor the spline in the baseline specification in Table 1. Positive health bias
(Ri>0) and negative health bias (Ri<0) are continuous health bias measures. For more information,
see Section 4.2 and 5.
54
Table A9: Relative Perception Bias and Risky Health Behaviors: Physical vs. Mental Health as Benchmark
(1) (2) (3) (4) (5) (6) (7) (8)
Physicial Health (SF12) Mental Health (SF12)
No sports Obese No healthy diet Smoker No sports Obese No healthy diet Smoker
Positive health bias (Ri>0)0.0022*** 0.0023*** 0.0006 -0.0002 -0.0002 -0.0006* 0.0005 0.0001
(0.0006) (0.0004) (0.0006) (0.0004) (0.0005) (0.0003) (0.0005) (0.0003)
Negative health bias (Ri<0)0.0008 -0.0000 0.0007 -0.0000 0.0002 0.0019** 0.0021** -0.0009
(0.0013) (0.0008) (0.0013) (0.0009) (0.0010) (0.0009) (0.0010) (0.0006)
R20.0252 0.1333 0.0459 0.0461 0.0168 0.0318 0.0477 0.0855
socio-demographics & education yes yes yes yes yes yes yes yes
employment char. & income yes yes yes yes yes yes yes yes
month FE yes yes yes yes yes yes yes yes
Source: Berlin Aging Study II (BASE-II); * p<0.1, ** p<0.05, *** p<0.01; standard errors in parentheses. The descriptive statistics are in the
Appendix (Table A2). The model is estimated by OLS with n=1,804 observations. Positive health bias (Ri>0) and negative health bias (Ri<0)
are continuous health bias measures, using the physical (columns (1) to (4)) and mental health (columns (5) to (8)) components of the SF12 to
measure Hi. For more information, see Sections 4.2 and 5.
55
... Jürges 2007; Lindeboom and Van Doorslaer 2004;Ziebarth 2018). For instance, respondents who overestimate their health are less likely to exercise and sleep enough and more prone to eat unhealthy foods, drink alcohol daily, and have higher body mass index (BMI) values (Arni et al. 2021). In this paper, we extend the latter analysis by using longitudinal data to assess the impact of health misperceptions on risky behaviours among middle-aged and older Chinese adults, a population characterized by prevalent high blood pressure whose leading mortality risk factors are alcohol consumption and smoking (Zhou et al. 2019). ...
... We define these perceptions based on the survey question 'Have you been diagnosed with hypertension or dyslipidemia or diabetes/high blood sugar by a doctor?' (1 = yes, 0 = no), designating self-perceived health as good if HBP i = 0, dyslipidemia i = 0, or diabetes i = 0 but identifying a positive health bias (health overconfidence) if g HBP i < HBP i , g dyslipidemia i < dyslipidemia i , or g diabetes i < diabetes i . More specifically, following Arni et al. (2021), we designate individual perceived health status as H i ¼ A i H i , where H i is individual health and A i denotes the individual's absolute health perception bias. A i =1 implies that the respondent has a correct perception of his or her own health; A i >1 represents a positive health perception bias (health overconfidence); and A i <1 represents a negative health perception bias (Arni et al. 2021). ...
... More specifically, following Arni et al. (2021), we designate individual perceived health status as H i ¼ A i H i , where H i is individual health and A i denotes the individual's absolute health perception bias. A i =1 implies that the respondent has a correct perception of his or her own health; A i >1 represents a positive health perception bias (health overconfidence); and A i <1 represents a negative health perception bias (Arni et al. 2021). ...
Article
Full-text available
Using data from the 2011 and 2015 China Health and Retirement Longitudinal Study, this paper analyzes the relation between health perception biases and risky health behaviors among adults aged 45 and older. We compare objective health outcomes (including hypertension, dyslipidemia and diabetes) with perceived conditions to assess absolute health perception biases and hypothesize that the role of biased health perceptions is a potential predictor for risky health behaviors. We provide evidence for the existence of positive absolute health perception biases and further document clear associations between health overconfidence and higher probabilities of alcohol consumption, overweight, and obesity, with the notable exception of smoking.
... This better health status invokes two channels: first, it provides satisfaction per se; and second, it enables individuals to enjoy life activities more fully. Arni, Dragone, Goette, and Ziebarth (2021) show that not only objective health but also perceived health can be a significant determinant of individual behavior. This observation is confirmed by Nie, Wang, and Sousa-Poza (2021) who show that health overconfidence is associated with more alcohol consumption, overweight and obesity among Chinese adults aged 45 and older. ...
... Moreover, everything else equal, those who overestimate their health would enjoy life activities more (in "normal" times). While Arni et al. (2021) focus on the consequence for risky health behaviors when considering psychological outcomes, the interdependence between health (be it objective or perceived) and enjoyable activities could amplify the effect that social activities have on perceived health and, ultimately, life satisfaction. However, the amplifying effect can also operate in the opposite direction. ...
... Following Arni et al. (2021), we define the objective relative position (the ranking) in the population health distribution as: ...
Article
The health risks of the current COVID-19 pandemic, together with the drastic mitigation measures taken in many affected nations, pose an obvious threat to public mental health. To assess predictors of poor mental health in the context of the COVID-19 pandemic, this study first implements survey-based measures of health perception biases among Chinese adults during the pandemic. Then, it analyzes their relation to three mental health outcomes: life satisfaction, happiness, and depression (as measured by the CES−D). We show that the health overconfidence displayed by approximately 30% of the survey respondents is a clear risk factor for mental health problems; it is a statistically significant predictor of depression and low levels of happiness and life satisfaction. We also document that these effects are stronger in regions that experienced higher numbers of confirmed COVID-19 cases and deaths. Our results also offer clear guidelines for the implementation of effective interventions to temper health overconfidence, particularly in uncontrollable situations like the COVID-19 pandemic.
... We extend this nascent but highly topical literature by investigating the role of biased health perception in explaining the (non)uptake of preventive actions among Europeans aged at least 50 years. Biased health perception that is, over-or underestimating one's own health, was shown to increase substantially with age (9); it is further relevant for the adoption of risky health behaviours and has serious consequences for health and well-being (10,11). In addition, incorrect beliefs about one's own health can affect one's perception of susceptibility to a disease and how severe that disease will be-which are important elements of preventive action according to the Health Belief Model (12). ...
... Thus far, the SHARE COVID-19 survey has been conducted only once, therefore enabling cross-sectional analyses only. SHARE data is ex ante harmonised, allowing us to analyse 13 European countries that participated in both the COVID-19 survey and in survey wave 5, namely, Belgium, Czechia, We operationalise health perception using a well-established measure that considers the difference between subjective and objective health (9,11,18). More specifically, we compare survey respondents' self-reported and tested ability to stand up from a chair. ...