Economics and Human Biology 46 (2022) 101134

Contents lists available at ScienceDirect

Economics and Human Biology

journal homepage: www.elsevier.com/locate/ehb

Does better education mitigate risky health behavior? A mendelian

randomization study
Jutta Viinikainen a, *, 1, Alex Bryson b, c, d, 2, Petri Böckerman a, d, e, 3, Jaana T. Kari a, 4,
Terho Lehtimäki f, g, h, i, 5, Olli Raitakari j, k, l, 6, Jorma Viikari m, n, 7, Jaakko Pehkonen a, 8
a
University of Jyväskylä, Jyväskylä University School of Business and Economics, Jyväskylä, Finland
b
University College London, Social Research Institute, London, United Kingdom
c
National Institute of Economic and Social Research, London, United Kingdom
d
IZA Institute of Labor Economics, Bonn, Germany
e
Labour Institute for Economic Research LABORE, Helsinki, Finland
f
Tampere University, Department of Clinical Chemistry, Tampere, Finland
g
Fimlab Laboratories, Tampere, Finland
h
Tampere University, Faculty of Medicine and Health Technology, Tampere, Finland
i
Tampere University, Finnish Cardiovascular Research Center, Tampere, Finland
j
University of Turku and Turku University Hospital, Centre for Population Health Research, Turku, Finland
k
University of Turku, Research Centre of Applied and Preventive Cardiovascular Medicine, Turku, Finland
l
Turku University Hospital, Department of Clinical Physiology and Nuclear Medicine, Turku, Finland
m
University of Turku, Department of Medicine, Turku, Finland
n
Turku University Hospital, Division of Medicine, Turku, Finland

A R T I C L E I N F O A B S T R A C T

JEL classification: Education and risky health behaviors are strongly negatively correlated. Education may affect health behaviors
I12 by enabling healthier choices through higher disposable income, increasing information about the harmful ef­
I20 fects of risky health behaviors, or altering time preferences. Alternatively, the observed negative correlation may
I26
stem from reverse causality or unobserved confounders. Based on the data from the Cardiovascular Risk in Young
Keywords: Finns Study linked to register-based information on educational attainment and family background, this paper
Education
identifies the causal effect of education on risky health behaviors. To examine causal effects, we used a genetic
Health behavior
Mendelian randomization
score as an instrument for years of education. We found that individuals with higher education allocated more
Smoking attention to healthy habits. In terms of health behaviors, highly educated people were less likely to smoke. Some
Diet model specifications also indicated that the highly educated consumed more fruit and vegetables, but the results
Abusive drinking were imprecise in this regard. No causal effect was found between education and abusive drinking. In brief,
inference based on genetic instruments showed that higher education leads to better choices in some but not all
dimensions of health behaviors.

* Correspondence to: Jyväskylä University School of Business and Economics, University of Jyväskylä, P.O. Box 35, FI-40014 Jyväskylä, Finland.
E-mail address: [email protected] (J. Viinikainen).
1
ORCID: 0000-0002-4252-3147
2
ORCID: 0000-0003-1529-2010
3
ORCID: 0000-0002-5372-2985
4
ORCID: 0000-0001-5205-7031
5
ORCID: 0000-0002-2555-4427
6
ORCID: 0000-0001-9365-3702
7
ORCID: 0000-0001-6452-010X
8
ORCID: 0000-0002-9684-7139

https://doi.org/10.1016/j.ehb.2022.101134
Received 1 March 2021; Received in revised form 14 March 2022; Accepted 15 March 2022
Available online 23 March 2022
1570-677X/© 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
J. Viinikainen et al. Economics and Human Biology 46 (2022) 101134

1. Introduction explanation for this relationship is that education leads to better health
because the highly educated have better access to healthcare services
Health is an important determinant of human well-being. Individuals and higher disposable incomes, thus enabling healthier lifestyle choices
with poorer health have lower life satisfaction and weaker labor market (e.g., Cutler and Lleras-Muney, 2006). Education may also increase the
attachment (Strine et al., 2008; Currie, 2009; Sánchez and Tassot, 2014; productive and allocative efficiencies of health production, leading to
Almond et al., 2018). At the societal level, the economic burden of poor better health (Grossman, 2006). Assuming that higher education in­
health is substantial, so much so that improving health and well-being is creases productive efficiency, the more educated are able to attain better
a key policy priority in most countries. The high economic and societal health from a given level of input. They may, for example, better un­
costs of adverse health conditions reinforce the need for policy measures derstand a doctor’s advice. According to the allocative efficiency hy­
that prevent or alleviate these problems. pothesis, in turn, more educated individuals are more likely to choose
Health disparities begin to develop early in life. Twin and genetic inputs that promote health than the less educated. Higher education has
epidemiological studies have revealed that observed variations in health also been linked to better health literacy skills (van der Heide et al.,
are partly explained by genetic makeup (see, e.g., Burton et al., 2005; 2013) and an increased willingness to delay gratification (Perez-Arce,
Polderman et al., 2015). In utero environments are also crucial, with 2017). Therefore, a better understanding of the consequences of various
studies showing that maternal exposure to the influenza pandemic in health behaviors (e.g., with regard to diet and smoking) and higher
1918 (Almond and Mazumder, 2005), prenatal malnutrition during the patience levels among highly educated individuals may support their
Dutch Hunger Winter (Roseboom et al., 2011), exposure to Ramadan ability to make healthy choices.
(Schoeps et al., 2018), and sudden reductions in air pollution due to Education may improve health, but causality may also run from
plant closures during the fetal period (Chay and Greenstone, 2003) all better health to higher educational attainment or, alternatively, com­
have long-lasting impacts on health. Individuals’ initial overall health mon factors such as parental background may explain the relationship.
stock depreciates with age, but health can be maintained and improved Studies using statistical methods for causal inference have been incon­
by making investments in health capital (Grossman, 1972). Such in­ clusive in terms of the link between education and health. In their meta-
vestments include the allocation of time for physical exercise, reading analysis, Hamad et al. (2018) focused on studies that utilized historical
health-related information, and the consumption of market goods, such changes in compulsory schooling laws as an exogenous source of vari­
as healthcare services and diets. In contrast, detrimental, risky health ation in education. The results indicated that education had mixed but
behaviors, such as excessive alcohol consumption, unhealthy diet, and largely beneficial effects on a range of health outcomes. However, Xue
heavy smoking, accelerate the depreciation of health capital. Thus, an et al. (2021) concluded that after correcting for publication bias, the
individual’s state of health partly reflects the resources they have allo­ effect of education on health outcomes was essentially zero. Further­
cated to health production. more, studies ignoring endogeneity tended to overstate the link between
Because health behaviors are important determinants of health, education and health. Thus, the evidence on the relationship between
policies that encourage individuals to engage in better health behaviors education and health is inconclusive.
are also likely to promote health. In this paper, we study the effect of Previous studies have also found a significant negative correlation
education on risky health behaviors. We focus on smoking, abusive between education and measures of risky health behavior. Among
alcohol consumption, diet (i.e., the consumption of fruits and vegeta­ highly educated individuals, the prevalence of smoking and binge
bles), and the level of attention toward healthy habits. Previous studies drinking has been found to be much lower, and higher education levels
on education and health behaviors aiming for causal inference have have also been shown to be related to better dietary choices (e.g.,
been mostly based on historical school reforms, which may limit their Cawley and Ruhm, 2011; Cutler and Lleras-Muney, 2010; De
external validity. To identify the causal impact of education on healthy Irala-Estevez et al., 2000). However, as in the case of education and
behaviors, we use a genetic score as an instrumental variable for edu­ health, a key challenge in such empirical studies is that the negative
cation. Studies that use genetic endowments to identify the link between correlation between education and risky health behaviors may also be
education and health behavior are still rare and, to the best of our driven by unobserved characteristics. For example, instead of being an
knowledge, they are all based on UK Biobank data (Gage et al., 2018; outcome of higher education, differences in time preferences may reflect
Davies et al., 2019; Rosoff et al., 2021; Sanderson et al., 2019; Zhou initial differences in attitudes toward postponed utility, hence affecting
et al., 2019). Our study extends this research by providing evidence both education choices and risky health behaviors (Farrell and Fuchs,
outside the UK context using data from another highly economically 1982). In addition, reverse causality may explain the correlation be­
developed country, Finland. Our data, which is rich in terms of cova­ tween education and health behaviors. For example, excessive drinking
riates, allowed us to focus on several closely connected measures of during the teenage years may lead to lower educational attainment in
health behaviors (attention toward healthy habits, smoking, abusive adulthood (Cook and Moore, 1994; Renna, 2007). If the association
drinking, and consumption of fruits and vegetables) that provide a more between education and health behaviors reflects omitted variables or
comprehensive picture of the effect of education on health behaviors.9 A reverse causality, this implies that investments in education would not
novelty of our empirical approach is that we formally evaluated the reduce poor health by altering or preventing unhealthy behaviors.
sensitivity of our estimates to the exclusion restriction of the Mendelian Empirical studies have typically found a negative association be­
randomization method, following an approach proposed by van Kip­ tween education and smoking status, but the results based on exogenous
persluis and Rietveld (2018). changes in schooling laws have been inconclusive, indicating either a
negative (Arendt, 2005; Jensen and Lleras-Muney, 2012; Jürges et al.,
2. Previous work 2011) or zero effect (e.g., Braakmann, 2011; Clark and Royer, 2013;
Dilmaghani, 2021; Kemptner et al., 2011; Li and Powdthavee, 2015).
Several studies have found that higher education is related to better Similarly, higher education has been related to higher alcohol con­
health (for a meta-analysis, see Furnee et al., 2008). A potential sumption (e.g., Cutler and Lleras-Muney, 2010), but this relationship
may not be causal (Braakmann, 2011; Clark and Royer, 2013; Fletcher,
2015). However, education may alter drinking patterns by reducing
9 binge drinking (Li and Powdthavee, 2015). The results concerning fruit
Obesity is potentially an important health-related outcome that may be
related to education. However, an earlier paper (Böckerman et al., 2017) using and vegetable consumption have also been inconclusive. Previous
the same data investigated the relationship between education and BMI/obesity studies have typically found a positive correlation between education
using the Mendelian randomization method. Hence, we do not use BMI as an and fruit and vegetable consumption (Braakmann, 2011; Clark and
outcome variable in this paper. Royer, 2013; Cutler and Lleras-Muney, 2010; Li and Powdthavee, 2015),

2
J. Viinikainen et al. Economics and Human Biology 46 (2022) 101134

but causal evidence suggests both positive (Li and Powdthavee, 2015)
and zero (Braakmann, 2011; Clark and Royer, 2013) effects.
̂ i + α ′ Xi + ε i
yhi = α0 + α1 educ 2 (3)
Another strand of instrumental variable studies has used genetic As discussed below, under suitable conditions, this method—called
scores as an instrument for education. This method, called Mendelian MR—identifies the local average treatment effect (LATE) on compliers
randomization (MR), uses the random inheritance of genetic material (i.e., those whose years of education were higher because of genetic
from one’s mother or father to isolate the causal impact of inherited inheritance) and avoids biases related to the OLS results (von Hinke
traits, such as the propensity for more education, on outcomes such as et al., 2016).
healthy behaviors. Findings from such MR studies have suggested that Earlier research has identified differences in health behaviors be­
higher education reduces smoking (Gage et al., 2018; Davies et al., 2019; tween females and males. For example, males have been shown to be
Sanderson et al., 2019) and may change drinking patterns. Higher ed­ more likely to smoke or engage in binge drinking than females (Cawley
ucation appeared to increase alcohol intake (Davies et al., 2019; Zhou and Ruhm, 2011). To account for this, we also estimated the models
et al., 2019) because of increased drinking frequency (Rosoff et al., separately for females and males.
2021; Zhou et al., 2019), but binge drinking was shown to be less
prevalent among the highly educated (Rosoff et al., 2021). Thus, the 3.2.2. MR assumptions for identification of causal effects
higher disposable income of more educated individuals may increase According to Mendel’s law of segregation (first law) and independent
their overall alcohol consumption, but a higher awareness of risks assortment (second law), alleles segregate randomly when they are
related to binge drinking or social norms may reduce the number of passed from one generation to the next, and each trait is inherited
drinks per drinking day. Rosoff et al. (2021) also investigated the rela­ independently from other traits at conception. In the MR method, this
tionship between education and alcohol dependency. They did not find a exogenous variation is used to identify the causal links between the
relationship between education and alcohol dependency measures, such exposure—in our case, education—and the outcome variable. The rapid
as the inability to cease drinking or the frequency of needing a morning decline in genome sequencing costs has led to an increase in the number
drink. of genome-wide association studies (GWASs) that attempt to find re­
lationships between genetic variants (a single nucleotide polymorphism,
3. Data and methods SNP) and expressed phenotypes. By summing up the number of SNPs
that have been associated with the phenotype of interest in a GWAS,
3.1. Data researchers have created PGSs that indicate a genetic risk of developing
a disease or some other trait. In addition to raw frequency PGSs,
The Cardiovascular Risk in Young Finns Study (YFS) is a longitudinal weighted PGSs—in which each SNP is weighted based on its effect size in
study of 3596 participants who were randomly chosen from five Finnish the GWAS—have also been used as instruments in the MR setting to
university regions in 1980 (Raitakari et al., 2008). The participants capture exogenous variation in the exposure variable.
represent six age cohorts (ages 3, 6, 9, 12, 15, and 18 years in 1980), and The MR estimator avoids biases related to the OLS estimator (von
since 1980, several follow-ups have been conducted. The YFS includes Hinke et al., 2016) under the four key assumptions as follows: 1) inde­
comprehensive information on genetic markers, that are required to pendence: the PGS is not associated with any confounder of the educa­
calculate the polygenic score (PGS), which we used as an instrument for tion–health behavior relationship—that is, the PGS is as good as
years of education. randomly assigned; 2) relevance: there is a nonzero effect of the in­
To obtain information on YFS participants’ educational attainment, strument on treatment—that is, the PGS is statistically significantly
the YFS was linked to the Finnish Longitudinal Employer-Employee Data associated with education (strong instrument/relevance assumption); 3)
(FLEED) of Statistics Finland using unique personal identifiers. The exclusion: the PGS affects health behavior only via its association with
matching was exact; that is, there were no misreported identification education; and 4) monotonicity: higher PGS values lead only to higher,
(ID) codes. Data on parental education was drawn from Statistics Fin­ not lower, educational attainment for each individual.
land’s Longitudinal Population Census (LPC) from 1980 and were linked The independence assumption could be violated because of popula­
to the YFS-FLEED using personal identifiers. tion stratification (allele frequencies differ between population sub­
groups), assortative mating (phenotypes affect partner selection), or
3.2. Methods dynastic effects (parental phenotypes directly affect offspring pheno­
types). All participants in the YFS were Caucasian, which minimized the
3.2.1. Statistical model possibility that systematic allele differences between ethnic subgroups
To replicate standard observational studies of the literature, we first biased the results. Furthermore, the features of the Finnish education
used an Ordinary Least Squares (OLS) estimation where a health system minimize the potential threat that dynastic effects violate the
behavior (yhi, where h refers to a health behavior and i to an individual) independence assumption: education is free of charge at all levels, and
was regressed on years of education (educ), and a vector of pre­ the role of private schools in the national education system, which are
determined control variables (X): cohort, sex, region of residence, and also tuition free, is minimal. To address potential assortative mating, we
parental education (Eq. 1). Because random sampling of YFS was con­ used a variable for parental education, which separated between neither
ducted at the individual level, there was no need to cluster standard parent, one parent, and both parents having attained university-level
errors. education. In this study, we also evaluated the independence assump­
tion by testing whether there were systematic differences in observed
(1) characteristics between individuals with different PGS levels by
′
yhi = β0 + β1 educi + β2 Xi + εi
comparing the observable variables by the median value of the instru­
To identify causal effects, we used the instrumental variables esti­
ment (von Hinke et al., 2016). To further assess the possibility that
mation method, which employed a PGS for years of education as an
observed characteristics are related to the education PGS, we performed
instrument for education. In the first stage, the variable reflecting years
a balance test by regressing the education PGS on exogenous variables (i.
of education was regressed on the instrument (PGS) and predetermined
e., family background), while simultaneously controlling for sex, age
controls (X; Eq. 2). In the second stage, the education variable in Eq. (1)
cohort, and regional fixed effects.
was replaced with the predicted values and estimated with OLS (Eq. 3):
The relevance assumption is satisfied if the PGS is robustly associated
̂ i = δ0 + δ1 PGSi + δ′ Xi
educ (2) with years of education and if this correlation is strong, the instrument is
2
considered to have a strong first stage. The association between the PGS

3
J. Viinikainen et al. Economics and Human Biology 46 (2022) 101134

SNPs and years of education was examined in a GWASs by Okbay et al. habits, 2) abusive drinking patterns, 3) smoking, and 4) consumption of
(2016a). In our main analysis, we utilized a PGS that was obtained using fruit and vegetables. These also constitute the key components of health
a significance threshold p < 0.01. The most important advantage of this behavior that have been analyzed in the literature. The amount of
instrument was its strength due to its inclusion of many SNPs (the actual attention participants paid to their health habits was determined based
number of SNPs is unknown). High instrument strength limits the pos­ on responses to the question, “To what extent do you focus on your
sibility for finite sample bias, which is expected to decrease as instru­ health habits?” The question was assessed on a five-point scale (1 = I
ment strength increases. In addition, weak instruments tend to increase hardly pay any attention; 5 = I pay a lot of attention). The respondents’
bias, stemming from violations of the other MR assumptions (McMartin abusive drinking patterns were assessed using 11 self-rated questions
and Conley, 2020). Staiger and Stock (1997) have suggested that the (see Appendix 1). The responses were given in yes/no format (0 = no;
minimum standard for a strong first stage is that the first-stage F sta­ 1 = yes), and the sum of the items was used to measure the severity of
tistics exceed the value of 10. We used this general rule to evaluate the abusive drinking behaviors. Long-term smoking was measured using
strong instrument assumption. pack years—that is, by multiplying the number of packs of cigarettes
The exclusion restriction requires that the PGS affects health be­ smoked per day by the years the person has smoked. Cigarette pack
haviors only via its association with education. This assumption could be years measure the cumulative lifetime consumption of cigarettes. For
violated if genetic variants related to education years also affect health example, a person has a 10-pack-year history of smoking if he or she has
behaviors either directly or through pathways other than education smoked one pack of cigarettes daily for 10 years. Information on the
(pleiotropy) or if they are in linkage disequilibrium (co-inherited) with consumption of fruits and vegetables was collected using a food fre­
SNPs affecting health behaviors via other pathways. Earlier studies quency questionnaire, which has been developed and validated by the
found that SNPs related to years of education may have also been Finnish National Institute for Health and Welfare (Paalanen et al.,
associated with noncognitive skills (Demange et al., 2021), mental 2006). The participants were asked to report the daily frequency and
health (Okbay et al., 2016a; Lam et al., 2019; Lee et al., 2019), cogni­ proportion size of selected food items over the past 12 months. Average
tion, body mass index (BMI), and height (Okbay et al., 2016a). To daily intake (in grams) of specific food groups was calculated using the
provide the complementary results that may take into account the bias National Food Consumption Database, Fineli. The consumption of fruits
stemming from pleiotropy related to these traits, we augmented our was based on five food items (citrus, apple, other fruits, fruit preserves,
baseline models with PGSs for the Big Five personality traits (openness, and berries), and the consumption of vegetables was based on eight
conscientiousness, extraversion, agreeableness, and neuroticism), items (root vegetables, lettuce, cabbage, fruit vegetables, mushrooms,
depressive symptoms, bipolar disorders, schizophrenia, childhood IQ, onions, vegetable preserves, beans, and peas). The group “fruit vegeta­
BMI, and height. bles” includes vegetables that are technically fruits—for example, avo­
As noted earlier, the advantage of the instrument we used in our cado, tomato, pepper, cucumber, pumpkin, and eggplant.
main analysis is its strength. However, the drawback of using an in­
strument with numerous SNPs is that the risk of pleiotropy may increase. 3.3.2. Endogenous variable: years of education
To address this issue, we also used an alternative PGS that included 74 Information on educational attainment in 2011 was drawn from the
SNPs, which were associated with years of education at p < 5 × 10− 8 in FLEED. The level of education was converted into years of education
Okbay et al. (2016a). For the 74 SNP PGSs, we had information on in­ using the official formula used by Statistics Finland (i.e., upper sec­
dividual SNPs that allowed us to perform Sargan’s test of overidentifying ondary education and postsecondary nontertiary education = 12 years;
restrictions to assess the validity of the 74 SNP PGSs. We also performed short-cycle tertiary education = 14 years; bachelor’s level = 16 years;
Sargan’s test using our baseline PGS and the PGS based on 74 SNPs as master’s level = 18 years; and doctoral level = 21 years). In 2011, the
instruments. Finally, we used the pleiotropy robust MR method van YFS participants were between 34 and 49 years old. However, a small
Kippersluis and Rietveld (2018) for sensitivity analysis to determine fraction (1.6% of n = 1719) were still studying, so we ascribed their
how strong the violation of the exclusion restriction would have to be for years of education based on the highest degree they had obtained until
the causal effect to be zero. that point.
Finally, the monotonicity assumption may be violated due to gene­
–environment interaction (G×E), which occurs when genetic pre­ 3.3.3. Instrumental variables
dispositions are expressed differently in different environments. For To identify the causal links between education and health behaviors,
example, tuition fees may prevent poor students with high genetic we used two alternative PGSs for years of education. In our main models,
predispositions toward education from obtaining higher levels of edu­ we used a PGS that was calculated as a weighted sum of the genotyped
cation. In Finland, tuition-free education and extensive state-provided risk alleles (or imputed allele dosages) associated with years of educa­
financial support to students at the university level substantially tion (p < 0.01) in a GWAS by Okbay et al. (2016a). The weights were
reduce the possibility that financial constraints would hamper the op­ based on the effect sizes of each SNP to years of education. In the
portunities for children from lower socioeconomic backgrounds to robustness tests, we used a weighted PGS associated with years of ed­
obtain higher levels of education. In addition, a geographically extensive ucation at a significance level of p < 5 × 10− 8 that included 74 SNPs
higher education network in Finland reduces regional disparities in (Okbay et al., 2016a). Further information on genotyping and calcula­
educational opportunities. However, to mitigate potential biases from tion of the PGSs is provided in Appendix 2.
G×E, we estimated a reduced-form model in which the outcome variable
was explained by the PGS, as suggested by VanderWeele et al. (2014). 3.3.4. Control variables
Because reduced-form models do not exploit information on the expo­ In all models, we used the following baseline controls: parental ed­
sure that may express itself differently depending on the environment, ucation (0 = neither parent, 1 = one parent, and 2 = both parents have
potential biases stemming from G×E are eliminated. The reduced-form completed university-level education by the year 1980, from the LPC);
estimates identify the effect of the exposure on the outcome but not the region of residence in 1980 (four indicator variables from the YFS:
the quantitative size of the effect of interest. Southern Finland, Eastern Finland, Western Finland, and Northern
Finland); and birth year (from the YFS).
3.3. Measures We also used the following PGSs in robustness analyses: the number
of cigarettes smoked per day (Furberg et al., 2010), the Big Five per­
3.3.1. Outcome variables sonality traits (de Moor et al., 2012), BMI (Locke et al., 2015), height
We focused on the following four health behavior measures that (van der Valk et al., 2015), childhood IQ (Benyamin et al., 2014),
were determined in 2011: 1) attention participants paid to their healthy schizophrenia (Schizophrenia Working Group of the Psychiatric

4
J. Viinikainen et al. Economics and Human Biology 46 (2022) 101134

Genomics Consortium, 2014), bipolar disorders (Ruderfer et al., 2014), (0.074 SD) increase in attention paid towards healthy habits, a 0.07
and depressive symptoms (Okbay et al., 2016b). point (0.026 SD) decrease in abusive alcohol consumption patterns,
0.404 fewer pack-years of tobacco use, and a 7.306 g (0.027 SD) in­
4. Results crease in fruit and vegetable consumption per day. The MR results
(Table 3, Column 2) supported the OLS findings that the link was causal
4.1. Descriptive analyses in the case of attention paid to healthy habits and smoking. The MR
point estimates were larger compared to the OLS estimates indicating
The average age of the participants in the sample (n = 1719) was that a one-year increase in years of education was associated with a
41.9 years in 2011, the share of females in the sample was 55.7%, and 0.147 point (0.155 SD) increase in attention paid towards healthy habits
13.0% of the participants had at least one university-educated parent. and a 0.820 pack-year reduction in tobacco consumption. The MR point
The bivariate analyses (Table 1) revealed that higher education (i.e., estimate for fruit and vegetable consumption also increased compared to
above-median years of education) was significantly associated with the corresponding OLS point estimate (αMR = 14.980 vs. βOLS = 7.306),
better health behaviors. We found that individuals with higher educa­ but the coefficient was imprecisely estimated, making the coefficient
tion reported that they allocated more attention to healthy habits, were only marginally (p < 0.10) significant. The MR results did not show a
less likely to smoke, displayed less abusive drinking patterns, and statistically significant link between education and abusive drinking
consumed more fruits and vegetables. patterns. The first-stage F statistics in the MR models exceeded the level
Table 2 compares the observed variables by the median value of the of 10, supporting the relevance assumption.
instrument. Consistent with the patterns in Table 1, individuals with a Table 3, Columns 3 and 4, reports the results separately for females
higher genetic propensity toward higher education were less likely to and males. Based on the MR findings, both females and males with
smoke and expressed that they paid more attention to healthy habits. higher education allocated more attention to healthy habits and smoked
Individuals with higher education PGSs also displayed less abusive less. The point estimates for fruit and vegetable consumption were also
drinking patterns and consumed more fruit and vegetables, but in these positive, but as in Column 2, they were imprecisely estimated and, thus,
cases, the differences between the low and high PGS groups were not statistically insignificant.
significant. The average age and the share of females did not differ be­
tween high and low PGS individuals, which was consistent with the
4.3. Robustness analyses
independence assumption of MR. Furthermore, the completed years of
education were higher among individuals whose PGS values exceeded
To assess the sensitivity of our findings, we first augmented our
the median level, which was consistent with the relevance assumption of
baseline model with PGSs for traits, which, according to the previous
the instrument.
literature, may share common genetic components with years of edu­
Table 2 further shows that the share of the participants with high
cation, namely the Big Five personality traits, depressive symptoms,
parental education was higher among high PGS individuals, and genetic
bipolar disorders, schizophrenia, childhood IQ, BMI, and height. The
variation was clustered between geographical areas. This highlighted
results (Appendix 4, Column 1) showed that these additional controls
the importance of controlling these variables to ensure that thereafter
had only a minimal effect on the point estimates.
the instrument would not correlate with the error term and, thus, the
Second, we estimated a reduced-form model, where health behaviors
exclusion restriction would not be violated. In addition, the average
were regressed on the PGS. The results from the reduced-form models
level of PGSs for the Big Five of conscientiousness, bipolar disorders,
(Appendix 4, Column 2) were consistent with the MR results, indicating
schizophrenia, and BMI differed between individuals with low and high
that individuals with a higher genetic predisposition toward higher
genetic propensities for higher education. The results from the balance
education paid greater attention to healthy habits and were less likely to
test where the education PGS was regressed on parental education,
smoke.
while simultaneously controlling for sex, age cohort, and regional fixed
Third, the results based on the 74 SNP PGSs (Appendix 4, Column 3)
effects, indicated that family background was significantly correlated
also indicated that individuals with higher education paid more atten­
with the PGS for education (Appendix 3). This highlighted the impor­
tion to healthy habits and smoked less. However, contrary to previous
tance of controlling for parental background in the analyses.
model specifications, the point estimate indicating the link between
education and fruit and vegetable consumption was negative, although
4.2. Main results imprecisely estimated (p > 0.950), with the first-stage F statistics below
the general rule of 10. Sargan’s overidentifying restrictions tests sup­
The OLS estimates (Table 3, Column 1) showed that higher education ported the null hypothesis that all 74 SNPs yielded the same MR estimate
was associated with paying greater attention to healthy habits, less (p > 0.478), thus lending support to the instrument’s validity. In addi­
abusive alcohol use, lower levels of tobacco consumption, and increased tion, when the 74 SNP PGSs and the PGSs used in our baseline models
consumption of fruit and vegetables. The results showed that a one-year were used as instruments, Sargan’s tests did not reject the null hy­
increase in years of education was associated with a 0.07 point pothesis of instrument exogeneity (regarding all outcome variables,

Table 1
Comparison of health behaviors by completed years of education.
All Above-median years of Below- median years of Difference t-statistics N
education education

Attention toward healthy habits (2011) 3.538 3.732 3.323 0.410 9.116 *** 1719
(0.947) (0.870) (0.981)
Abusive alcohol consumption (scale: 0–11; 2011) 3.147 2.920 3.403 -0.483 -3.689 *** 1655
(2.666) (2.554) (2.766)
Smoking (pack years; 2011) 1.445 (5.391) 0.499 2.720 -2.221 -7.444 *** 1620
(2.581) (7.516)
Fruit and vegetable consumption (g/day; 2011) 448.554 472.191 421.902 50.289 3.509 *** 1468
(275.130) (259.213) (289.922)

Notes: The table reports the means and standard deviations in parentheses. The differences between groups were tested using a two-sample t-test. Statistically sig­
nificant at the * 10%, ** 5%, and *** 1% levels.

5
J. Viinikainen et al. Economics and Human Biology 46 (2022) 101134

Table 2
Comparison of observables by the instrument value.
All Above-median PGS Below-median PGS Difference t-statistics N

Attention toward healthy habits (2011) 3.538 3.639 3.437 0.202 4.446 *** 1719
(0.947) (0.930) (0.953)
Abusive alcohol consumption (scale 0–11; 2011) 3.147 3.131 3.163 -0.032 -0.245 1655
(2.666) (2.677) (2.656)
Smoking (pack years, 2011) 1.445 0.929 2.006 -1.077 -3.991 *** 1620
(5.391) (4.604) (6.086)
Fruit and vegetable consumption (g/day; 2011) 448.554 452.471 444.572 7.899 0.550 1468
(275.130) (266.542) (283.721)
Female (share) 0.557 0.556 0.558 -0.002 -0.070 1719
(0.497) (0.497) (0.497)
Age (2011) 41.904 41.848 41.960 -0.113 -0.465 1719
(5.040) (5.126) (4.954)
Additional PGSs
Openness -6.698 -6.437 -6.959 0.522 1.038 1719
(10.422) (10.724) (10.111)
Conscientiousness -1.536 -0.927 -2.144 -1.217 2.201 ** 1719
(11.474) (11.509) (11.474)
Extraversion 1.158 1.288 1.029 0.259 0.476 1719
(11.282) (11.589) (10.973)
Agreeableness 37.763 37.930 37.597 0.332 0.726 1719
(9.494) (9.457) (9.534)
Neuroticism -6.012 -6.170 -5.855 -0.315 -0.458 1719
(14.261) (13.893) (14.627)
Depressive symptoms 0.029 0.022 0.035 -0.014 -0.463 1719
(0.610) (0.607) (0.614)
Bipolar disorders -8.552 -8.350 -8.754 0.405 2.648 *** 1719
(3.172) (3.135) (3.199)
Schizophrenia -4.302 -4.427 -4.178 -0.249 -1.754 * 1719
(2.944) (2.926) (2.958)
Childhood IQ 0.977 1.017 0.937 0.080 1.528 1719
(1.091) (1.097) (1.085)
BMI 0.968 0.907 1.029 -0.122 -3.479 *** 1719
(0.729) (0.707) (0.745) (0.035)
Height 2.123 2.214 2.032 0.182 1.466 1719
(2.569) (2.568) (2.567)
Region of residence (1980)
- Southern Finland 0.166 0.182 0.151 0.030 1.695 * 1719
(0.373) (0.386) (0.358)
- Western Finland 0.362 0.364 0.359 0.005 0.219 1719
(0.481) (0.482) (0.480)
- Eastern Finland 0.310 0.290 0.330 -0.040 -1.810 * 1719
(0.463) (0.454) (0.471)
- Northern Finland 0.162 0.164 0.159 0.005 0.273 1719
(0.368) (0.371) (0.366)
Education years (2011) 13.994 14.746 13.242 1.504 11.888 *** 1719
(2.728) (2.812) (2.420)
Parental 0.181 0.244 0.117 0.127 5.312 *** 1719
education (scale 0–2; 1980) (0.500) (0.566) (0.414)

Notes: The table reports the means and standard deviations in parentheses. The differences between groups were tested using a two-sample t-test. Parental education
variable equals 0 if neither parent, 1 if one parent, and 2 if both parents had university-level education. Statistically significant at the * 10%, ** 5%, and *** 1% levels.

p > 0.206). 5. Discussion

Fourth, we augmented the model for tobacco consumption with a
PGS for cigarettes smoked per day. Controlling for a genetic predispo­ Using longitudinal data that combined survey information on health
sition toward smoking had only a very minor effect on the point estimate behaviors with register-based information on education, we found that
for years of education (baseline MR model, Table 3, Column 2: − 0.820, higher education was related to better health behaviors in terms of
p < 0.01; augmented model: − 0.816, p < 0.01). YFS data did not paying greater attention to healthy habits, less abusive drinking pat­
contain a PGS for other outcome variables. terns, lower levels of tobacco consumption, and higher amounts of fruit
As a final robustness check, we followed van Kippersluis and Rietveld and vegetable consumption. The MR results, which identified the causal
(2018) approach to assess the sensitivity of our results to the violation of effect of education on health behaviors, showed that this link was causal
the exclusion restriction. We found that any violation of the exclusion in the case of attention paid to healthy habits and smoking. The MR
restriction would have to account for 43–44% of the first stage effect in specifications implied that education increased fruit and vegetable
order to reduce the education point estimates to zero in relation to intake, but these coefficients were imprecisely estimated. The MR results
healthy habits and smoking. In the case of fruit and vegetable con­ did not reveal a statistically significant link between education and
sumption, violation of the exclusion restriction would have to account abusive alcohol consumption.
for 19% of the variance in the first stage to reduce the causal effect to Previous studies have shown a strong link between low education
zero. These results are shown and discussed in Appendix 5. and a higher incidence of risky health behaviors, such as smoking, binge
drinking, and dietary choices (e.g., De Irala-Estevez et al., 2000; Cawley
and Ruhm, 2011). Our OLS estimates were consistent with these results.
Previous studies that used historical changes in schooling laws to

6
J. Viinikainen et al. Economics and Human Biology 46 (2022) 101134

Table 3 sensitive to providing socially desirable answers to questions related to

Education and health behaviors. Results based on linear regression (OLS) and health behaviors, which may have biased our results. Second, MR relies
Mendelian randomization (MR) methods. on the assumption that the instrument is valid. A potential threat to the
(1) OLS (2) MR (3) MR, (4) MR, validity assumption is pleiotropy—that is, that the SNPs in the PGSs
female male affected health behaviors through pathways other than education. To
Attention toward 0.070 *** 0.147 *** 0.150 *** 0.139 *** evaluate the sensitivity of results to this issue we followed van Kipper­
healthy habits (0.009) (0.030) (0.038) (0.046) sluis and Rietveld (2018) approach to determine how strong the viola­
[0.053; [0.089; [0.075; [0.050; tion of the exclusion restriction would have to be for the causal effect to
0.087] 0.205] 0.225] 0.229]
be zero. We found that in the case of fruit and vegetable consumption, a
First-stage F . 162.20 103.71 64.16
statistics relatively mild violation of the exclusion would reduce the causal esti­
R2 0.093 . . . mate to zero. Regarding smoking and attention to healthy habits, the
N 1719 1719 958 761 results were less sensitive, but the possibility of zero effects also applied
Abusive alcohol -0.070 *** -0.043 -0.092 0.031 to these findings. Furthermore, our analyses indicated that the education
consumption (0.025) (0.083) (0.104) (0.132)
[− 0.119; [− 0.206; [− 0.297; [− 0.228;
PGS was associated with other PGSs that may also affect health behav­
− 0.022] 0.120] 0.112] 0.290] iors. However, controlling for these additional PGSs had only a minor
First-stage F . 155.13 96.41 64.01 effect on the point estimates. It is not possible to prove instrument val­
statistics idity in an MR or, in general, in any instrumental variable study, but the
R2 0.069 . . .
robustness tests we performed were consistent with instrument validity.
N 1655 1655 918 737
Smoking (pack -0.404 *** -0.820 *** -0.586 ** -1.069 *** Third, the LATE that the MR model identified captured the average
years) (0.052) (0.197) (0.240) (0.311) treatment effect among those whose years of education were higher
[− 0.506; [− 1.206; [− 1.056; [− 1.677; because of genetic inheritance. Variation in years of education driven by
− 0.301] − 0.435] − 0.116] − 0.460] other factors, for example, changes in schooling laws, could have
First-stage F . 153.09 88.32 69.79
different impacts on health behaviors. Fourth, in all but one case
statistics
R2 0.068 . . . (abusive alcohol consumption), the MR estimates suggested a stronger
N 1620 1620 913 707 link between education and health behaviors than the OLS estimates.
Fruit and 7.306 *** 14.980 * 10.261 18.882 This difference may stem from two sources. First, there may have been
vegetable (2.751) (9.108) (13.029) (12.003)
unobserved confounders that biased the OLS estimate downward. Sec­
consumption [1.911; [− 2.872 [− 15.275; [− 4.644;
(g/day) 12.702] 32.832] 35.797] 42.407] ond, although OLS identified the average treatment effect (ATE) if ed­
First-stage F . 129.05 70.26 62.16 ucation was uncorrelated with the error term, the LATE, which the MR
statistics method identified, captured the ATE among compilers—that is, those
R2 0.062 . . . whose years of education increased via the impact of our genetic in­
N 1468 1468 848 620
strument. Thus, these two methods may capture different treatment
Notes: The table reports regression coefficients, their standard errors (in parameters in a setting where there are heterogeneous treatment effects.
parenthesis), and 95% confidence intervals (in square brackets).The instrument Our findings suggest that the benefits of education are not restricted
used in the MR models is the polygenic score for education years based on ge­ to better labor market outcomes, but that education also promotes less
netic markers (p < 0.01). The unit of analysis is the individual. The six cohorts risky health behaviors. Finland is a Nordic welfare state with relatively
under study are drawn from the Young Finns study. The models include (unre­
low differences in health outcomes (e.g., life expectancy) between
ported) controls for indicators for age, sex (Columns 1 and 2, being female),
different socioeconomic groups (OECD, 2019) and a highly ranked
region of residence in 1980, and parental education in 1980. Statistically sig­
nificant at the * 10%, ** 5%, and *** 1% levels. comprehensive education system financed by the government, a system
that is tuition-free for all students (OECD, 2020). Our findings suggest
that even in a country where social equality has been a key policy pri­
identify causal effects arrived at mixed conclusions regarding the
ority for many decades, education may lead to meaningful differences in
effectiveness of education on improving health behaviors. Typically,
health behaviors and, thus, health outcomes in the long term. This im­
these studies have been less optimistic about the effectiveness of
plies that public and health policies should pay particular attention to
schooling than association studies reporting correlations between the
low-educated individuals and to their investments in health capital. If
variables of interest. Although some studies have found that education
socioeconomic differences in health behaviors accumulate to large dif­
decreased smoking (Arendt, 2005; Jensen and Lleras-Muney, 2012;
ferences in health outcomes in the long run, the health care system may
Jürges et al., 2011) and (binge) drinking (Li and Powdthavee, 2015), as
not be able to mitigate these differences.
well as increased consumption of fruit and vegetables (Li and Powd­
thavee, 2015), many studies have found no effect of education on these
health behaviors (Braakmann, 2011; Clark and Royer, 2013; Dilma­ 6. Conclusions
ghani, 2021; Kemptner et al., 2011; Fletcher, 2015). Studies that have
used MR as an identification strategy have found that higher education In our study, we showed that higher education led to better choices
reduced smoking (Gage et al., 2018; Davies et al., 2019; Sanderson et al., in terms of some dimensions of health behaviors. Thus, the benefits of
2019) and increased alcohol intake (Davies et al., 2019; Zhou et al., higher education are not only restricted to monetary outcomes (i.e.,
2019) because education increased drinking frequency (Rosoff et al., higher employment, lower unemployment, and higher earnings), which
2021; Zhou et al., 2019). Our results provided support for the view that have been extensively documented in the literature (e.g., Card, 1999),
education reduces smoking and possibly increases fruit and vegetable but they also apply to nonmonetary domains of human well-being.
consumption. In accordance with Rosoff et al. (2021), we did not find a
significant link between education and abusive alcohol consumption. Funding
Four issues should be considered when interpreting the results. First,
it is possible that individuals with higher education levels are more The Young Finns Study has been financially supported by the
Academy of Finland: grant numbers 322098, 286284, 134309 (Eye),

7
J. Viinikainen et al. Economics and Human Biology 46 (2022) 101134

126925, 121584, 124282, 129378 (Salve), 117787 (Gendi), 41071 CRediT authorship contribution statement
(Skidi); the Social Insurance Institution of Finland; Competitive State
Research Financing of the Expert Responsibility area of Kuopio; Tam­ Jutta Viinikainen Conceptualization, Methodology, Formal anal­
pere and Turku University Hospitals (grant number X51001); Juho ysis, Writing – original draft, Writing – review & editing; Alex Bryson
Vainio Foundation; Paavo Nurmi Foundation; Finnish Foundation for Conceptualization, Methodology, Writing – review & editing; Petri
Cardiovascular Research; Finnish Cultural Foundation; the Sigrid Juse­ Böckerman Conceptualization, Methodology, Writing – review & edit­
lius Foundation, Finland; Tampere Tuberculosis Foundation, Finland; ing; Jaana T. Kari Conceptualization, Writing – review & editing; Terho
Emil Aaltonen Foundation; Yrjö Jahnsson Foundation; Signe and Ane Lehtimäki Conceptualization, Writing – review & editing; Olli Raita­
Gyllenberg Foundation; Jenny and Antti Wihuri Foundation; Diabetes kari Conceptualization, Writing – review & editing, Funding acquisi­
Research Foundation of Finnish Diabetes Association; EU Horizon 2020, tion, Project administration (YFS); Jorma Viikari Conceptualization,
Belgium (grant number 755320 for TAX-INOMISIS); European Research Writing – review & editing; Jaakko Pehkonen Conceptualization,
Council, Belgium (grant number 742927 for MULTIEPIGEN project); Writing – review & editing, Funding acquisition, Project administration
Tampere University Hospital Supporting Foundation; and the Society of (YFS-FLEED-LPC).
Finnish Clinical Chemistry. The use of the YFS-FLEED-LPC data has been
supported by Palkansaajasäätiö and OP Group Research Foundation. Declarations of interest

None.

Appendix 1. Items indicating abusive alcohol consumption

Share of “yes” answers

Have you ever had the habit of drinking alcohol before going to a party? 56.80%
Do you have a habit of drinking a bottle of wine or a similar amount of beer or other alcoholic drinks over the weekend? 43.63%
Have you ever had a daily habit of drinking a small amount of alcohol to relax? 24.77%
Have you ever felt that you need to drink more alcohol than before to have the same effect? 16.86%
Have you ever had trouble drinking less than your friends? 8.76%
Have you ever fallen asleep after drinking a reasonable amount of alcohol without knowing how you went to bed? 24.47%
Have you ever had a bad conscience after drinking alcohol? 51.90%
Have you ever taken a drink to cure a hangover? 23.26%
Have you ever tried to avoid drinking for a certain period, for example, a week? 34.14%
Have you ever found it difficult to stop drinking once you have started? 13.60%
Has a relative or a friend been concerned about your drinking or suggested you cut down? 16.50%
The average number of “yes” answers 3.147
Notes: The questions were answered on a yes/no scale (0 = no; 1 = yes). The sum of items indicates the severity of excessive drinking behavior. n = 1655.

APPENDIX 2. Genotyping and calculation of the PGS

Genotyping in the YFS was implemented using the Illumina Bead Chip (Human 670 K) from 2442 YFS participants, including 546,677 SNPs, and
the genotypes were called using the Illumina clustering algorithm (Teo et al., 2007). Quality control was performed using the Sanger genotyping QC
pipeline, and individuals with possible relatedness were removed. SHAPEIT v1 and IMPUTE 2 software (Delaneau et al., 2012) were used for genotype
imputation with the 1000 Genomes Phase I Integrated Release Version 3 (March 2012 haplotypes) as a reference panel (Howie et al., 2009; 1000
Genomes Project Consortium, 2010).
GeneticRiskScoreCalculator version 0.1.0c was used for calculating several PGSs in parallel. Independent effect SNPs for each summary-statistic
file were identified by double-clumping, first using a 350-kb distance and then a 5-Mb distance with R2 = 0.1 as a linkage-disequilibrium threshold. A
weighted PGS was calculated by summing risk alleles for each independent SNP using its GWAS effect size (β or log (OR) from the GWAS study) as a
weight. Five GWAS p-value thresholds (p < 5 × 10− 8, 1 × 10− 5, 1 × 10− 4, 1 × 10− 3, and 1 × 10− 2) were used for constructing PGSs for each
summary-statistic file. The human leukocyte antigen region (chr6:25,000,000–35,000,000) was omitted from the calculations. The PGSs were scaled
between 0 and 2 for compatibility with the QTL-mapping pipeline (Võsa et al., 2021).

Appendix 3. Balance test

Coefficient (SE) 95% confidence interval

Parental education 0.171 *** (0.021) 0.129; 0.212

Notes: Table reports the OLS results from regression models where the instrument for education
was regressed on parental education, while simultaneously controlling for sex, age cohort, and
regional fixed effects. Statistically significant at the * 10%, * *5%, and * ** 1% levels.
n = 2412.

8
J. Viinikainen et al. Economics and Human Biology 46 (2022) 101134

Appendix 4. Education and health behaviors: robustness analyses

(1) (2) (3)

MR OLS MR
Additional PGS controls Reduced-form model Using 74 SNP PGS

Attention toward healthy habits 0.141 * ** 0.214 * ** 0.199 * *

(0.031) (0.043) (0.101)
[0.081; 0.201] [0.130; 0.298] [− 0.000; 0.398]
First-stage F statistics 151.18 . 15.73
R2 . 0.071 .
N 1719 1719 1719
Abusive alcohol consumption -0.056 -0.062 0.015
(0.086) (0.120) (0.249)
[− 0.225; 0.112] [− 0.297; 0.174] [− 0.473; 0.503]
First-stage F statistics 146.67 . 16.73
R2 . 0.064 .
N 1655 1655 1655
Smoking (pack years) -0.830 * ** -1.193 * ** -1.538 * *
(0.204) (0.289) (0.711)
[− 1.229; − 0.430] [− 1.760; − 0.625] [− 2.930; − 0.145]
First-stage F statistics 144.52 . 11.21
R2 . 0.043 .
N 1620 1620 1620
Fruit and vegetable consumption (g/day) 15.851 * 20.076 -2.117
(9.326) (12.306) (38.188)
[− 2.427; 34.129] [− 4.063; 44.215] [− 76.964; 72.729]
First-stage F statistics 119.12 . 8.55
R2 . 0.059 .
N 1468 1468 1468
Notes: The table reports regression coefficients, their standard errors (in parenthesis), and 95% confidence intervals (in square brackets). The reported MR
coefficients for years of education are obtained from models which use the polygenic score (PGS) for years of education as an instrument. Column 1 utilizes a
PGS that was obtained using a significance threshold p < 0.01 and Column 2 a PGS based on 74 SNPs. The models in Column 1 also include additional controls
for PGSs for the Big Five personality traits, depressive symptoms, bipolar disorders, schizophrenia, childhood IQ, BMI, and height. Column 2 shows the OLS
coefficients for PGS (p < 0.01) for years of education. The unit of analysis is the individual. The six cohorts under study are drawn from the Young Finns study.
The models include (unreported) controls for indicators for age, sex, region of residence in 1980, and parental education in 1980. Statistically significant at the
* 10%, * *5%, and * ** 1% levels.

Appendix 5. Pleiotropy-robust Mendelian randomization

The plausibly exogenous estimation method (Conley et al., 2012) is an instrumental variables approach that can be applied when the instrument
validity is debatable. It provides a method for performing inference in a situation where the exclusion restriction does not necessarily hold pre­
cisely—i.e., the instrument is plausibly or approximately exogenous. van Kippersluis and Rietveld (2018) applied this method to the MR setting
(pleiotropy-robust Mendelian randomization, PRMR) and used it as a sensitivity analysis to determine how strong the violation of the exclusion
restriction would have to be for the causal effect to be zero.
In their example, van Kippersluis and Rietveld (2018) estimated the effect of educational attainment on BMI. We followed their approach (and
STATA codes) to address the potential horizontal pleiotropy problem. Fig. A1 plots the results. Alcohol consumption was omitted from the analyses
because our MR results did not imply a causal link between education and excessive alcohol consumption. In Fig. A1, lambda refers to the percentage
of standardized effect of the PGS on years of education, which is the direct effect of the PGS on standardized health behavior outcomes. The estimate
for lambda = 0 corresponds to the MR point estimate. Moving to the right on the x-axis implies a stronger violation of the exclusion restriction. The
results showed that the causal effect of education on attention to healthy habits was estimated to be zero when lambda = 0.43. Thus, a violation of the
exclusion restriction 43% of the first stage effect would decrease the point estimate to zero. Regarding smoking and fruit and vegetable consumption,
the corresponding lambdas were 0.44 and 0.19, respectively. Furthermore, the 95% confidence intervals included zeros at the following levels of
lambda: attention to healthy habits: 0.2, smoking: 0.2, and fruit and vegetable consumption: 0. In their study, van Kippersluis and Rietveld (2018)
interpreted that “a relative mild violation” of the exclusion restriction of 29% based on the confidence interval produced “at best weak evidence” of
causal effects. Our PRMR results regarding attention to healthy habits and smoking were comparable to those of van Kippersluis and Rietveld (2018),
implying that it is important to consider the possibility that zero effect cannot be ruled out. In the case of fruit and vegetable consumption, even a
milder violation of the exclusion restriction would reduce the causal effect to zero.

9
J. Viinikainen et al. Economics and Human Biology 46 (2022) 101134

Fig. A1. The causal effect of education on healthy habits for varying values of lambda. Note: The gray area indicates the 95% confidence interval.

References Davies, N.M., Hill, W.D., Anderson, E.L., Sanderson, E., Deary, I.J., Smith, G.D., 2019.
Multivariable two-sample Mendelian randomization estimates of the effects of
intelligence and education on health. Elife 8, e43990.
1000 Genomes Project Consortium, 2010. A map of human genome variation from
De Irala-Estevez, J., Growth, M., Johansson, L., Oltersdorf, U., Prättälä, R., Martínez-
population-scale sequencing. Nature 467 (7319), 1061–1073.
González, M.A., 2000. A systematic review of socio-economic differences in food
Almond, D., Mazumder, B., 2005. The 1918 influenza pandemic and subsequent health
habits in Europe: consumption of fruit and vegetables. Eur. J. Clin. Nutr. 54 (9), 706.
outcomes: an analysis of SIPP data. Am. Econ. Rev. 95 (2), 258–262.
de Moor, M.H., Costa, P.T., Terracciano, A., et al., 2012. Meta-analysis of genome-wide
Almond, D., Currie, J., Duque, V., 2018. Childhood circumstances and adult outcomes:
association studies for personality. Mol. Psychiatry 17 (3), 337–349.
Act II. J. Econ. Lit. 56 (4), 1360–1446.
Delaneau, O., Marchini, J., Zagury, J.F., 2012. A linear complexity phasing method for
Arendt, J.N., 2005. Education effects on health: a panel data analysis using school reform
thousands of genomes. Nat. Methods 9 (2), 179–181.
for identification. Econ. Educ. Rev. 24 (2), 149–160.
Demange, P., Malanchini, M., Mallard, T., Biroli, P., 2021. Investigating the genetic
Benyamin, B., Pourcain, B., Davis, O.S., et al., 2014. Childhood intelligence is heritable,
architecture of non-cognitive skills using GWAS-by-subtraction. Net. Genet. 53,
highly polygenic and associated with FNBP1L. Mol. Psychiatry 19 (2), 253–258.
35–44.
Böckerman, P., Viinikainen, J., Pulkki-Råback, L., Hakulinen, C., Pitkänen, N.,
Dilmaghani, M., 2021. Education, smoking and health: evidence from Canada. Educ.
Lehtimäki, T., Pehkonen, J., Raitakari, O.T., 2017. Does higher education protect
Econ. 29 (5), 490–508.
against obesity? Evidence using Mendelian randomization. Prev. Med. 101,
Farrell, P., Fuchs, V.R., 1982. Schooling and health: the cigarette connection. J. Health
195–198.
Econ. 1 (3), 217–230.
Braakmann, N., 2011. The causal relationship between education, health and health
Fletcher, J.M., 2015. New evidence of the effects of education on health in the US:
related behaviour: evidence from a natural experiment in England. J. Health Econ.
Compulsory schooling laws revisited. Soc. Sci. Med. 127, 101–107.
30 (4), 753–763.
Furberg, H., Kim, Y., Dackor, J., et al., 2010. Genome-wide meta-analyses identify
Burton, P.R., Tobin, M.D., Hopper, J.L., 2005. Key concepts in genetic epidemiology.
multiple loci associated with smoking behavior. Nat. Genet. 42 (5), 441–447.
Lancet 366 (9489), 941–951.
Furnee, C.A., Groot, W., van den Brink, H., 2008. The health effects of education: a meta-
Card, D., 1999. The causal effect of education on earnings. In: Ashenfelter, O.C., Card, D.
analysis. Eur. J. Public Health 18 (4), 417–421.
(Eds.), Handbook of Labor Economics, Vol. 3A. Elsevier, pp. 1801–1863.
Gage, S.H., Bowden, J., Davey Smith, G., Munafo, M.R., 2018. Investigating causality in
Cawley, J., Ruhm, C.J., 2011. The economics of risky health behaviors. In: Pauly, M.V.,
associations between education and smoking: a two-sample Mendelian
McGuire, T.G., Barros, P.P. (Eds.), Handbook of Health Economics, Vol. 2. Elsevier,
randomization study. Int. J. Epidemiol. 47 (4), 1131–1140.
pp. 95–199.
Grossman, M., 1972. On the concept of health capital and the demand for health. J. Polit.
Chay, K.Y., Greenstone, M., 2003. The impact of air pollution on infant mortality:
Econ. 80 (2), 223–255.
evidence from geographic variation in pollution shocks induced by a recession. Q. J.
Grossman, M., 2006. Education and nonmarket outcomes. In: Hanushek, E., Welch, F.
Econ. 118 (3), 1121–1167.
(Eds.), Handbook of the Economics of Education, first ed., Vol 1. Elsevier/North
Clark, D., Royer, H., 2013. The effect of education on adult mortality and health:
Holland, pp. 577–633.
Evidence from Britain. Am. Econ. Rev. 103 (6), 2087–2120.
Hamad, R., Elserb, H., Tranc, D.C., Rehkopfc, D.H., Goodman, S.N., 2018. How and why
Conley, T.G., Hansen, C.B., Rossi, P.E., 2012. Plausibly exogenous. Rev. Econ. Stat. 94
studies disagree about the effects of education on health: a systematic review and
(1), 260–272.
meta-analysis of studies of compulsory schooling laws. Soc. Sci. Med. 212, 168–178.
Cook, P.J., Moore, M.J., 1994. This tax’s for you: the case for higher beer taxes. Natl. Tax.
von Hinke, S., Smith, G.D., Lawlor, D.A., Propper, C., Windmeijer, F., 2016. Genetic
J. 47 (3), 559–573.
markers as instrumental variables. J. Health Econ. 45, 131–148.
Currie, J., 2009. Healthy, wealthy, and wise: socioeconomic status, poor health in
Howie, B.N., Donnelly, P., Marchini, J., 2009. A flexible and accurate genotype
childhood, and human capital development. J. Econ. Lit. 47 (1), 87–122.
imputation method for the next generation of genome-wide association studies. PLoS
Cutler, D.M., Lleras-Muney, A., 2006. Education and health: evaluating theories and
Genet. 5 (6), e1000529.
evidence. Natl. Bur. Econ. Res. Work. Pap. 12352.
Jensen, R., Lleras-Muney, A., 2012. Does staying in school (and not working) prevent
Cutler, D.M., Lleras-Muney, A., 2010. Understanding differences in health behaviors by
teen smoking and drinking? J. Health Econ. 31 (4), 644–657.
education. J. Health Econ. 29 (1), 1–28.

10
J. Viinikainen et al. Economics and Human Biology 46 (2022) 101134

Jürges, H., Reinhold, S., Salm, M., 2011. Does schooling affect health behavior? Evidence Ruderfer, D.M., Fanous, A.H., Ripke, S., et al., 2014. Polygenic dissection of diagnosis
from the educational expansion in Western Germany. Econ. Educ. Rev. 30 (5), and clinical dimensions of bipolar disorder and schizophrenia. Mol. Psychiatry 19
862–872. (9), 1017–1024.
Kemptner, D., Jürges, H., Reinhold, S., 2011. Changes in compulsory schooling and the Sánchez, A.C., Tassot, C., 2014. An exploration of the determinants of the subjective
causal effect of education on health: evidence from Germany. J. Health Econ. 30 (2), well-being of Americans during the great recession. OECD Economics Department
340–354. Working Papers, No. 1158. OECD Publishing, Paris. https://doi.org/10.1787/
Lam, M., Hill, W.D., Trampush, J.W., et al., 2019. Pleiotropic meta-analysis of cognition, 5jxzc9p57lq3-en. Accessed 1 November 2021.
education, and schizophrenia differentiates roles of early neurodevelopmental and Sanderson, E., Smith, G.D., Bowden, J., Munafò, M.R., 2019. Mendelian randomisation
adult synaptic pathways. Am. J. Hum. Genet. 105 (2), 334–350. analysis of the effect of educational attainment and cognitive ability on smoking
Lee, P.H., Anttila, V., Won, H., et al., 2019. Genomic relationships, novel loci, and behaviour. Nat. Commun. 10 (1), 1–9.
pleiotropic mechanisms across eight psychiatric disorders. Cell 179 (7), 1469–1482. Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014. Biological
Li, J., Powdthavee, N., 2015. Does more education lead to better health habits? Evidence insights from 108 schizophrenia-associated genetic loci. Nature 511 (7510),
from the school reforms in Australia. Soc. Sci. Med. 127, 83–91. 421–427.
Locke, A.E., Kahali, B., Berndt, S.I., et al., 2015. Genetic studies of body mass index yield Schoeps, A., van Ewijk, R., Kynast-Wolf, G., Nebié, E., Zabré, P., Sié, A., Gabrysch, S.,
new insights for obesity biology. Nature 518 (7538), 197–206. 2018. Ramadan exposure in utero and child mortality in Burkina Faso: analysis of a
McMartin, A., Conley, D., 2020. Commentary: Mendelian randomization and population-based cohort including 41,025 children. Am. J. Epidemiol. 187 (10),
education–Challenges remain. Int. J. Epidemiol. 49 (4), 1193–1206. 2085–2092.
OECD, 2019. Health at a glance 2019: OECD Indicators. OECD Publishing, Paris. Staiger, D., Stock, J.H., 1997. Instrumental variables regression with weak instruments.
OECD, 2020. Education policy outlook: Finland. OECD Publishing, Paris (Accessed 1 Econometrica 65, 447–486.
November 2021). 〈http://www.oecd.org/education/policy-outlook/country-profile Strine, T.W., Chapman, D.P., Balluz, L.S., Moriarty, D.G., Mokdad, A.H., 2008. The
-Finland-2020.pdf〉. associations between life satisfaction and health-related quality of life, chronic
Okbay, A., Beauchamp, J.P., Fontana, M.A., et al., 2016a. Genome-wide association illness, and health behaviors among US community-dwelling adults. J. Community
study identifies 74 loci associated with educational attainment. Nature 533 (7604), Health 33 (1), 40–50.
539–542. Teo, Y.Y., Inouye, M., Small, K.S., Gwilliam, R., Deloukas, P., Kwiatkowski, D.P.,
Okbay, A., Baselmans, B.M., De Neve, J.E., et al., 2016b. Genetic variants associated with Clark, T.G., 2007. A genotype calling algorithm for the Illumina BeadArray platform.
subjective well-being, depressive symptoms, and neuroticism identified through Bioinformatics 23 (20), 2741–2746.
genome-wide analyses. Nat. Genet. 48 (6), 624–633. van der Heide, I., Wang, J., Droomers, M., Spreeuwenberg, P., Rademakers, J., Uiters, E.,
Paalanen, L., Männistö, S., Virtanen, M.J., Knekt, P., Räsänen, L., Montonen, J., 2013. The relationship between health, education, and health literacy: results from
Pietinen, P., 2006. Validity of a food frequency questionnaire varied by age and body the Dutch Adult Literacy and Life Skills Survey. J. Health Commun. 18 (sup1),
mass index. J. Clin. Epidemiol. 59 (9), 994–1001. 172–184.
Perez-Arce, F., 2017. The effect of education on time preferences. Econ. Educ. Rev. 56, van der Valk, R.J., Kreiner-Møller, E., Kooijman, M.N., et al., 2015. A novel common
52–64. variant in DCST2 is associated with length in early life and height in adulthood.
Polderman, T.J., Benyamin, B., De Leeuw, C.A., Sullivan, P.F., van Bochoven, A., Hum. Mol. Genet 24 (4), 1155–1168.
Visscher, P.M., Posthum, D., 2015. Meta-analysis of the heritability of human traits van Kippersluis, H., Rietveld, C.A., 2018. Pleiotropy-robust Mendelian randomization.
based on fifty years of twin studies. Nat. Genet. 47 (7), 702. Int. J. Epidemiol. 47 (4), 1279–1288.
Raitakari, O.T., Juonala, M., Rönnemaa, T., et al., 2008. Cohort profile: the VanderWeele, T.J., Tchetgen, E.J.T., Cornelis, M., Kraft, P., 2014. Methodological
cardiovascular risk in Young Finns Study. Int. J. Epidemiol. 37 (6), 1220–1226. challenges in Mendelian randomization. Epidemiology 25 (3), 427.
Renna, F., 2007. The economic cost of teen drinking: late graduation and lowered Võsa, U., et al., 2021. Large-scale cis- and trans-eQTL analyses identify thousands of
earnings. Health Econ. 16 (4), 407–419. genetic loci and polygenic scores that regulate blood gene expression. Nat. Genet 53
Roseboom, T.J., Painter, R.C., van Abeelen, A.F., Veenendaal, M.V., de Rooij, S.R., 2011. (9), 1300–1310.
Hungry in the womb: what are the consequences? Lessons from the Dutch famine. Xue, X., Cheng, M., Zhang, W., 2021. Does education really improve health? A meta-
Maturitas 70 (2), 141–145. analysis. J. Econ. Surv. 35 (1), 71–105.
Rosoff, D.B., Clarke, T.K., Adams, M.J., McIntosh, A.M., Smith, G.D., Jung, J., Lohoff, F. Zhou, T., Sun, D., Li, X., Ma, H., Heianza, Y., Qi, L., 2019. Educational attainment and
W., 2021. Educational attainment impacts drinking behaviors and risk for alcohol drinking behaviors: mendelian randomization study in UK Biobank. Mol. Psychiatry
dependence: results from a two-sample Mendelian randomization study with~ 26, 4355–4366.
780,000 participants. Mol. Psychiatry 26, 1119–1132.

11