Positive affect during adolescence and health and well-being in adulthood: An outcome-wide longitudinal approach

Abstract

Background
Several intergovernmental organizations, including the World Health Organization and United Nations, are urging countries to use well-being indicators for policymaking. This trend, coupled with increasing recognition that positive affect is beneficial for health/well-being, opens new avenues for intervening on positive affect to improve outcomes. However, it remains unclear if positive affect in adolescence shapes health/well-being in adulthood. We examined if increases in positive affect during adolescence were associated with better health/well-being in adulthood across 41 outcomes.

Methods and findings
We conducted a longitudinal cohort study using data from Add Health—a prospective and nationally representative cohort of community-dwelling U.S. adolescents. Using regression models, we evaluated if increases in positive affect over 1 year (between Wave I; 1994 to 1995 and Wave II; 1995 to 1996) were associated with better health/well-being 11.37 years later (in Wave IV; 2008; N = 11,040) or 20.64 years later (in Wave V; 2016 to 2018; N = 9,003). Participants were aged 15.28 years at study onset, and aged 28.17 or 37.20 years—during the final assessment. Participants with the highest (versus lowest) positive affect had better outcomes on 3 (of 13) physical health outcomes (e.g., higher cognition (β = 0·12, 95% CI = 0·05, 0·19, p = 0.002)), 3 (of 9) health behavior outcomes (e.g., lower physical inactivity (RR = 0·80, CI = 0·66, 0·98, p = 0.029)), 6 (of 7) mental health outcomes (e.g., lower anxiety (RR = 0·81, CI = 0·71, 0·93, p = 0.003)), 2 (of 3) psychological well-being (e.g., higher optimism (β = 0·20, 95% CI = 0·12, 0·28, p < 0.001)), 4 (of 7) social outcomes (e.g., lower loneliness (β = −0·09, 95% CI = −0·16, −0·02, p = 0.015)), and 1 (of 2) civic/prosocial outcomes (e.g., more voting (RR = 1·25, 95% CI = 1·16, 1·36, p < 0.001)). Study limitations include potential unmeasured confounding and reverse causality.

Conclusions
Enhanced positive affect during adolescence is linked with a range of improved health/well-being outcomes in adulthood. These findings suggest the promise of testing scalable positive affect interventions and policies to more definitively assess their impact on outcomes.

Full text

RESEARCH ARTICLE
Positive affect during adolescence and health
and well-being in adulthood: An outcome-
wide longitudinal approach
Eric S. KimID
1,2,3‡
*, Renae WilkinsonID
2‡
, Sakurako S. OkuzonoID
4
, Ying ChenID
2,5
,
Koichiro Shiba
6
, Richard G. Cowden
2
, Tyler J. VanderWeele
2,5,7
1Department of Psychology, University of British Columbia, Vancouver, British Columbia, Canada, 2Human
Flourishing Program, Institute for Quantitative Social Science, Harvard University, Cambridge,
Massachusetts, United States of America, 3Lee Kum Sheung Center for Health and Happiness, Harvard T.
H. Chan School of Public Health, Boston, Massachusetts, United States of America, 4Department of Social
& Behavioral Sciences, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, United States of
America, 5Department of Epidemiology, Harvard T.H. Chan School of Public Health, Boston,
Massachusetts, United States of America, 6Department of Epidemiology, Boston University School of Public
Health, Boston, Massachusetts, United States of America, 7Department of Biostatistics, Harvard T.H. Chan
School of Public Health, Boston, Massachusetts, United States of America
‡ ESK and RW share first authorship on this work.
*eric.kim@ubc.ca
Abstract
Background
AU :Pleaseconfirmthatallheadinglevelsarerepresentedcorrectly:Several intergovernmental organizations, including the World Health Organization and
United Nations, are urging countries to use well-being indicators for policymaking. This
trend, coupled with increasing recognition that positive affect is beneficial for health/well-
being, opens new avenues for intervening on positive affect to improve outcomes. However,
it remains unclear if positive affect in adolescence shapes health/well-being in adulthood.
We examined if increasesAU :PerPLOSstyle;italicsshouldnotbeusedforemphasis:Hence;allitalicizedwordshavebeenchangedtoregulartextthroughoutthearticle:in positive affect during adolescence were associated with better
health/well-being in adulthood across 41 outcomes.
Methods and findings
We conducted a longitudinal cohort study using data from Add Health—a prospective and
nationally representative cohort of community-dwelling U.S. adolescents. Using regression
models, we evaluated if increases in positive affect over 1 year (between Wave I; 1994 to
1995 and Wave II; 1995 to 1996) were associated with better health/well-being 11.37 years
later (in Wave IV; 2008; N= 11,040) or 20.64 years later (in Wave V; 2016 to 2018; N=
9,003). Participants were aged 15.28 years at study onset, and aged 28.17 or 37.20 years—
during the final assessment. Participants with the highest (versus lowest) positive affect had
better outcomes on 3 (of 13) physical health outcomes (e.g., higher cognition (β= 012, 95%
CI = 005, 019, p= 0.002)), 3 (of 9) health behavior outcomes (e.g., lower physical inactivity
(RR = 080, CI = 066, 098, p= 0.029)), 6 (of 7) mental health outcomes (e.g., lower anxiety
(RR = 081, CI = 071, 093, p = 0.003)), 2 (of 3) psychological well-being (e.g., higher opti-
mism (β=020, 95% CI = 012, 028, p<0.001)), 4 (of 7) social outcomes (e.g., lower
PLOS MEDICINE
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 1 / 17
a1111111111
a1111111111
a1111111111
a1111111111
a1111111111
OPEN ACCESS
Citation: Kim ES, Wilkinson R, Okuzono SS, Chen
Y, Shiba K, Cowden RG, et al. (2024) Positive affect
during adolescence and health and well-being in
adulthood: An outcome-wide longitudinal
approach. PLoS Med 21(4): e1004365. https://doi.
org/10.1371/journal.pmed.1004365
Received: July 3, 2023
Accepted: February 22, 2024
Published: April 2, 2024
Copyright: ©2024 Kim et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: Parts of the datasets
generated and/or analyzed during the current study
are publicly available in the Adolescent to Adult
Health repository (https://www.cpc.unc.edu/
projects/addhealth/documentation/publicdata).
However, this study utilized the extensive
restricted-use data available by contractual
agreement. Per the Add Health website (https://
data.cpc.unc.edu/projects/2/view), "Restricted-Use
Data will be distributed only to certified researchers
who commit themselves to maintaining limited
access. To be eligible to enter into a contract,
researchers must complete Contract Application
which includes: Security plan IRB approval letter

loneliness (β=−009, 95% CI = −016, −002, p= 0.015)), and 1 (of 2) civic/prosocial out-
comes (e.g., more voting (RR =125, 95% CI = 116, 136, p<0.001)). Study limitations
include potential unmeasured confounding and reverse causality.
Conclusions
Enhanced positive affect during adolescence is linked with a range of improved health/well-
being outcomes in adulthood. These findings suggest the promise of testing scalable posi-
tive affect interventions and policies to more definitively assess their impact on outcomes.
Author summary
Why was this study done?
• Intergovernmental organizations, including the World Health Organization and United
Nations, are advocating for the inclusion of well-being indicators in policy-making,
alongside traditional economic measures like GDP.
• This trend, along with the growing understanding of positive affect’s benefits for health
and well-being, creates exciting opportunities for intervening on positive affect to
enhance outcomes, but, existing research primarily focuses on adult populations, and
although some progress has been made, it remains unclear if positive affect assessed in
adolescence shapes health/well-being in adulthood.
• Thus, there is a need to understand this knowledge gap to help inform relevant policies
and interventions.
What did the researchers do and find?
• In our longitudinal cohort study of U.S. adolescents from the Add Health project, we
examined if increases in positive affect over 1 year during adolescence were associated
with better health/well-being outcomes on 41 indicators in adulthood.
• In our analyses, we used data from 11,040 participants in Wave IV and 9,003 partici-
pants in Wave V, prioritizing the latest available data from Wave V whenever possible,
and including Wave IV data when Wave V data was unavailable.
• The study, which spanned from adolescence (age 15.28) to adulthood (ages 28.17 or
37.20), found that higher positive affect is generally associated with better outcomes in
various areas, including physical health, health behavior, mental health, psychological
well-being, and social and civic engagement.
What do these findings mean?
• Our findings indicate that fostering positive affect during adolescence could lead to a
broad spectrum of improved health and well-being outcomes in adulthood, underscor-
ing its potential value in youth-focused policies and interventions.
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 2 / 17
$1000 payment by check (NEW contract only).
This website also has links for how to apply for the
restricted-use dataset. No third-party data was
used.” Code used to run analyses for this study are
available at the Open Science Framework (please
see: https://osf.io/uz7sr/).
Funding: This work was supported by grants from
the Michael Smith Health Research BC (https://
healthresearchbc.ca/; ESK), Canadian Institutes of
Health Research (https://cihr-irsc.gc.ca/; ESK), and
the John Templeton Foundation (https://www.
templeton.org/; TJV) The funders did not play any
role in the study design, data collection and
analysis, decision to publish, or preparation of the
manuscript.
Competing interests: E.S.K. has worked as a
consultant with AARP and UnitedHealth Group. T.J.
V. reports receiving licensing fees from Flerish Inc.
and Flourishing Metrics.
Abbreviations: ADHD, AU :Anabbreviationlisthasbeencompiledforthoseusedinthetext:Pleaseverifythatallentriesarecorrect:attention deficit
hyperactivity disorder; CES-D, Center for
Epidemiological Studies Depression; CI, confidence
interval; NNH, Number Needed to Harm; NNT,
Number Needed to Treat; OR, odds ratio; PPVT,
Peabody Picture Vocabulary Test; PTSD,
posttraumatic stress disorder; RR, relative risk;
STI, sexually transmitted infection.

• The implications for practice include the potential development of targeted strategies to
enhance positive affect among adolescents as a means to promote long-term health and
well-being.
• However, limitations such as possible unmeasured confounding factors, possible reverse
causality, and reliance on self-reported data suggest that these findings should be
approached as preliminary, highlighting the need for further research in this area.
Introduction
Three factors converge to highlight a unique opportunity for improving the health/well-being
of adolescents and adults. First, several prominent intergovernmental organizations (e.g.,
Organization for Economic Co-operation and Development, World Health Organization,
United Nations) are urging nations to use well-being indicators (e.g., happiness), in addition
to traditional economic indicators (e.g., gross domestic product), when sculpting policies [1].
Many countries are adopting this paradigm shift. Second, according to the Lancet Commission
on Adolescent Health and Well-being, “Failure to invest in the health of the largest generation
of adolescents in the world’s history jeopardises earlier investments in maternal and child
health, erodes future quality and length of life, and escalates suffering, inequality, and social
instability” [2]. For this effort, identifying factors that foster health/well-being among adoles-
cents is crucial. While much effort has focused on identifying risk factors of disease, investiga-
tors are increasingly seeking potentially modifiable health assets [3–7]. Targeting health assets
during adolescence, a critical developmental phase for acquiring health assets and establishing
healthy behaviors and mindsets, is a promising point of intervention that can enhance the tra-
jectory of health/well-being across the life course. Third, positive affect—the experience of
pleasurable emotions, such as happiness, joy, excitement, enthusiasm, calm, and contentment
—is one promising health asset, and emerging research shows that it is associated with a range
of health/well-being outcomes [8]. Although some progress has been made, it remains unclear
if positive affect assessed in adolescence shapes a wide range of health/well-being outcomes in
adulthood. Additionally, positive affect has been declining in younger populations over the
past decade and this trend is an ever-growing concern [9]. Thus, as the number of govern-
ments focusing on well-being grows, so do opportunities to target positive affect for its own
sake, as well as a means to enhance the health/well-being of adolescents and adults.
Positive affect is shaped by social structures and changing life circumstances [10], it is also
modifiable through various interventions that can be applied among individuals (e.g., therapy,
online exercises, physical activity) [11] and at a national scale (e.g., policies) [12]. Accumulat-
ing research conducted in adults indicates that high positive affect is linked to improved health
behaviors (e.g., increased: medication adherence, physical activity, sleep, diet), enhanced bio-
logical function (e.g., healthier: immune function, inflammation levels, lipid levels), and
decreased risk of chronic diseases (e.g., stroke, cardiovascular disease) and mortality [3,8,13].
A small number of landmark studies have prospectively evaluated how positive affect (and
related constructs) in adolescents might influence subsequent health outcomes. For example,
in adolescents, positive affect and psychological well-being composite scores (composed of ele-
ments such as positive affect, life satisfaction, purpose in life, optimism, etc.) have been linked
with decreased risk of poor health outcomes (e.g., healthier cardiometabolic profiles, healthier
BMI) [14–16]. However, evidence for associations with health behaviors are mixed. For
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 3 / 17

example, positive affect has been linked with improved health behaviors (e.g., higher physical
activity, healthier diets, not smoking, lower composite health behaviors index scores composed
of factors like: physical inactivity, fast food consumption, binge drinking, smoking) [14,17,18],
but there have also been interesting null results (e.g., no associations with: smoking, maintain-
ing physical activity or becoming physically active, maintaining health diets, sleep duration)
[14,19].
When considering prior studies conducted in younger populations, they have contributed
substantially to the literature. However, they remain limited in some ways. First, many are
cross-sectional, making it difficult to evaluate causality. Second, many health behavior and
physical health outcomes have not yet been examined. Third, numerous studies do not
account for key potential confounders (e.g., depressive symptoms/negative affect or baseline
physical health). Fourth, many studies evaluated composite measures of psychological well-
being (making it difficult to determine if specific aspects of psychological functioning are
“active ingredients”) and/or outcomes (making it difficult to determine which specific out-
comes are driving composite outcome scores). Fifth, we are unaware of longitudinal studies
that analyzed data in a way that allows researchers to ask a different question of particular
importance in this era of translational research: What health/well-being outcomes might we
observe if positive affect were intervened upon?
To begin addressing this question, we used an outcome-wide analytic approach [20]. This is
a hypothesis-generating, data-driven analytic approach aimed at discovering estimates of the
outcomes we might expect to observe if positive affect was intervened upon. Promising find-
ings can then undergo further investigation in future studies. We leveraged a large, prospec-
tive, and nationally representative sample of U.S. adolescents, and examined if increases in
positive affect over 1 year during adolescence were associated with better subsequent health/
well-being 11.37 years or 20.64 years later across 41 health/well-being outcomes.
Methods
Study population
We used data from Add Health, a prospective and nationally representative sample of U.S.
adolescents in grades 7 to 12 during the 1994 to 1995 school year (Wave I), and this sample
was followed into adulthood. Using a stratified random sampling approach, 80 high schools
and 52 feeder middle schools were selected because they collectively capture a representative
sample of U.S. adolescents in terms of ethnicity, urbanicity, school size, school type, and region
of country. Among the 20,745 students who participated in Wave I in-home interviews, we
excluded adolescents who did not: participate in the Wave II survey (in which the exposure
variable positive affect was assessed, n= 6,009, year 1995 to 1996) or have valid survey weights
at the respective outcome wave (Wave IV n= 3,698, year 2008; Wave V n= 5,733, year 2016 to
2018), resulting in a final analytic sample of N= 11,040 for Wave IV outcomes and N= 9,003
for Wave V outcomes (see Figs A and B in S1 Appendix for additional details). The reason we
have 2 outcome waves is because we used data from the latest available wave, which was Wave
V, whenever possible. However, 10 important outcomes were not assessed in Wave V, but
assessed in Wave IV. In these cases, we used data from Wave IV.
Our study used data from 3 time points (t
0
, t
1
, and t
2
). Almost all covariates were assessed
in the pre-baseline wave (t
0
, Wave I—when Add Health participants were in grades 7 to 12
and aged 11 to 21). However, 3 covariates, not assessed during Wave I, were retrospectively
assessed in Wave IV. This choice was made because controlling for potential confounders in
the pre-baseline wave establishes a clear temporal order between covariates, exposure, and out-
comes that helps alleviate the risk of adjusting for a potential mediator [20]. The exposure,
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 4 / 17

positive affect, was then assessed 1 year later in the baseline wave (t
1
, Wave II—when Add
Health participants were adolescents and aged 12 to 21). All outcomes were assessed another
11.73 or 20.64 years later in the outcome waves (t
3
, Wave IV—when Add Health participants
were early adults aged 24 to 32 years; t
3
, Wave V—when participants were early-midlife adults
aged 33 to 43), depending on data availability. We recognize that defining adolescence, the
phase of life that stretches between childhood and adulthood, has long posed a challenge. The
definition of adolescence as 10 to 19 years of age originates from the mid-20th century, a
period when patterns of adolescent growth and the timing of role transitions were starkly dif-
ferent than modern times. Thus, an expanded and more inclusive definition of adolescence
with an upper limit of 21 (i.e., American Academy of Pediatrics and the US Department of
Health and Human Services) [21] or even 24 (i.e., Lancet Commission on Adolescent Health
and Wellbeing) [22] aligns more closely with contemporary patterns of adolescent biological
growth and social role transitions. Additionally, only 1.1% of our study sample (n= 118) was
outside the ages of 10 to 19 at study baseline.
Add Health provides extensive documentation about their protocol, instrumentation, and
complex sampling strategy elsewhere (https://addhealth.cpc.unc.edu/). Add Health has been
approved by several ethics committees, including the University of North Carolina IRB. Fur-
ther, informed consent was obtained from all respondents.
Measures
Positive affect. The exposure was assessed at baseline (t
1
; Wave II) and also at pre-base-
line (t
0
; Wave I) using the positive affect subscale of the Center for Epidemiological Studies
Depression (CES-D) Scale—[23] a subscale that repeatedly emerges in factor analytic studies
as illustrated in meta-analyses [24] and has been used repeatedly in past research [25]. Using a
four-point scale (range 1 to 4), respondents rated the degree to which they experienced the fol-
lowing items in the past week: “I was happy,” “I felt hopeful about the future,” “I felt that I was
just as good as other people,” “I enjoyed life.” We averaged all responses and created a compos-
ite score so that higher scores reflect higher positive affect (range 1 to 4). The Cronbach’s α
coefficient, which assesses the internal consistency reliability of the scale, was 0.73. To examine
potential threshold effects, we created tertiles based on the distribution of positive affect scores
in the sample.
Covariates. Covariates were assessed in the pre-baseline wave (t
0
; Wave I, the closest wave
before the exposure assessment), unless otherwise noted, and included: sociodemographic and
family factors (age, sex, race/ethnicity (White, Black, Hispanic, Asian, Other), born in the U.S.,
geographic region (Northeast, Midwest, South, West), two-parent household, number of sib-
lings, household income quintile, household welfare receipt, health insurance, smoker in
household, mother age, mother race/ethnicity, parents born in the U.S., parental education
(<high school, high school, some college, college), mother employed full time, mother reli-
gious service attendance (never or seldom, <1×/week, 1×/week), mother self-rated health,
mother happiness, parent has a disability, parent has obesity, parent has alcoholism, childhood
maltreatment by parents (assessed retrospectively at Wave IV), psychosocial and academic fac-
tors (including mental health condition diagnosis [assessed retrospectively at Wave IV], nega-
tive affect, self-esteem, life expectancy, parental control, relationship quality with parent,
religious service attendance, has romantic partner, has a learning disability, cognitive develop-
ment [assessed with Peabody Picture Vocabulary Test [PPVT]], school connectedness, GPA,
delinquency), and health status and health behaviors (somatic symptoms, pubertal develop-
ment range, physical health condition diagnosis [assessed retrospectively at Wave IV], over-
weight/obesity, functional limitations, self-rated health, suicidal ideation, sleep disturbance,
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 5 / 17

physical inactivity, cigarette smoking, binge drinking, marijuana use, illicit drug use, history of
STIs, preventative health care use).
Outcomes. We considered 41 outcomes which were assessed in the outcome waves (t
2
;
Wave IV or Wave V), and include dimensions of: physical health (number of diagnosed physi-
cal health conditions, cancer, high cholesterol, hypertension, diabetes, asthma, sleep apnea,
migraines, allostatic load, overweight/obesity, functional limitations, cognition, self-rated
health), health behavior (sleep disturbance, physical inactivity, cigarette smoking, binge drink-
ing, marijuana use, prescription drug misuse, illicit drug use, history of sexually transmitted
infections [STIs], preventative health care use), mental health (depression diagnosis, anxiety
diagnosis, posttraumatic stress disorder [PTSD] diagnosis, attention deficit hyperactivity dis-
order [ADD/ADHD] diagnosis, negative affect, suicidal ideation, perceived stress), psychologi-
cal well-being (optimism, job satisfaction, sense of control), social factors (relationship quality
with parent, social activities, social support, loneliness, romantic relationship quality, satisfac-
tion with parenting, perceived discrimination), and civic and prosocial behavior (voting, vol-
unteering). A full description of each outcome can be found in Text A in S1 Appendix. These
outcomes were chosen because they are frequently included in the conceptualization of key
models that characterize the antecedents, processes, and outcomes that foster positive adoles-
cent and adult development [26–28].
Statistical analysis
We used an outcome-wide analytic approach [20], which has several characteristics not widely
used outside of biostatistics and causal inference. Thus, we summarize those characteristics
here. First, we run a separate model for each outcome and consistently use the same set of
covariates across all models, and all outcomes. Second, we control for covariates in the wave
prior to the exposure since, if we assess potential confounders in the same time point as the
exposure (t
1
), it remains unclear if they are confounders or mediators; if we accidentally con-
trol for mediators in the same time point, we may spuriously attenuate true effects. A prag-
matic approach to avoiding this problem is by adjusting for potential confounders in the pre-
baseline wave (t
0
). Third, to enhance our ability to strive toward “no unmeasured confound-
ing,” and “exchangeability” (as well as other criteria described in “disjunctive cause criterion”
for selection of covariates that includes potential causes of either the exposure or the outcomes
or both), which all enhance our ability to make causal inference, we adjust for a sufficiently
rich set of potential confounder variables to make these assumptions plausible [29,30]. Fourth,
to reduce potential reverse causality we also adjust for all outcome variables in the pre-baseline
wave (t
0
). Fifth, to evaluate potential “change” in positive affect we adjust for positive affect in
the pre-baseline wave (t
0
). This helps “hold constant” pre-baseline levels of positive affect (see
Text B in S1 Appendix for proof). Adjusting for pre-baseline levels of positive affect (t
0
) also
has several other advantages including helping reduce risk of reverse causality and also
“removing” the accumulating effects positive affect already had on outcomes in the past
(“prevalent exposure”) and allowing readers to instead focus on the effects of change in posi-
tive affect (“incident exposure”) over 1 year, on outcomes.
Separate models were run for each outcome. Depending on the nature of the outcome, a
different model was run: (1) for each binary outcome with a prevalence <10%, logistic regres-
sion was used; (2) for each binary outcome with a prevalence 10%, generalized linear model
(with a log link and Poisson distribution) was used; and (3) for each continuous outcome, a
linear regression model was used. Further, each continuous outcome was standardized
(mean = 0, SD = 1). All analyses were weighted to account for unequal probability of selection
and attrition and adjust for the complex sampling design.
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 6 / 17

In our results section, we comment on the traditional 0.05 p-value threshold and provide
95% confidence intervals for all effect estimates, which are often considered preferable assess-
ments of uncertainty since all thresholds are ultimately arbitrary. The study is reported accord-
ing to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE)
guideline (Checklist A in S1 Appendix). This study did not have a formally registered prospec-
tive protocol or analysis plan. However, all analyses were discussed and planned before we
began analyses.
Secondary analyses. We carried out several additional analyses. First, we performed sen-
sitivity analysis using E-values to evaluate the exposure–outcome association’s robustness to
unmeasured confounding [31]. Second, we reanalyzed all models using only complete cases to
evaluate the potential impact of using multiple imputation for handling missing data. Third, to
provide a baseline comparison, we reanalyzed all models without any control for potential
confounders. Fourth, we reanalyzed all models using a reduced list of potential confounders
more conventionally used in the social/behavioral sciences (i.e., sociodemographic factors) to
evaluate how similar (or different) our results were to past research. Fifth, we calculated the
Number Needed to Treat (NNT) and Number Needed to Harm (NNH) for each outcome.
Multiple imputation. In our dataset, the percentage of missing data varied across vari-
ables and ranged from 0% to 26% (see Table A in S1 Appendix). We imputed missing data for
the covariates and outcomes using an imputation by chained equations procedure by generat-
ing 5 datasets. This method provides a more flexible approach than other methods of handling
missing data [32]. All analyses were conducted in Stata 17.
Results
In our study, the final cohort consisted of 11,040 participants for Wave IV outcomes and 9,003
participants for Wave V outcomes. At the pre-baseline wave, when the potential confounders
were assessed, participants were 15 years old (SD = 160), and just over half were women
(n= 5,901; 53.45%). Participants reported being White (n= 5,960; 54.02%), Black (n= 2,253;
20.42%), Hispanic (n= 1,745; 15.82%), Asian (n= 675; 6.12%), and “Other” (n= 400; 3.63%).
The distribution of sociodemographic and health characteristics at pre-baseline was generally
consistent across positive affect tertiles, but there were some key differences: those in the high-
est (versus lowest) positive affect tertile had higher household income (e.g., 23% versus 14%
were in the highest income quintile) and had a higher percentage of two-parent households
(75% versus 64%). Table 1 describes the distribution of covariates. Table B in S1 Appendix
shows how positive affect changed from the pre-baseline to baseline wave. The average length
of follow-up from baseline (Wave II) to the outcome waves was 11.37 years (for Wave IV out-
comes; SD = 0.50, range 10 to 12) or 20.64 years (for Wave V outcomes; SD = 0.71, range = 19
to 22).
Table 2 shows the associations between positive affect and subsequent health/well-being
outcomes. When considering physical health outcomes and health behaviors, positive affect
was associated with healthier functioning, or score, on 3 (out of 13) physical health outcomes,
including lower likelihood of migraines (relative risk [RR]=079, 95% confidence interval
[CI] = 067, 093, p= 0.005), as well has higher cognition (β=012, 95% CI = 005, 019,
p= 0.002), and self-rated health (β=011, 95% CI = 005, 018, p<0.001). However, there was
little evidence of an association with number of diagnosed physical health conditions, any of
the other specific physical health conditions (i.e., cancer, high cholesterol, hypertension, diabe-
tes, asthma, and sleep apnea), allostatic load, overweight/obesity, or functional limitations.
When considering health behaviors, positive affect was associated with healthier functioning,
or scores, on 3 (out of 9) health behavior outcomes, including: lower likelihood of prescription
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 7 / 17

Table 1. Characteristics of participants at pre-baseline by tertiles of baseline positive affect (National Longitudinal Study of Adolescent to Adult Health [Add
Health])AU :Abbreviationlistshavebeencompiled=updatedforthoseusedinTables1to3:Pleaseverifythatallentriesarecorrect:.
Positive affect
Characteristic Tertile 1 (n= 4,184) Tertile 2 (n= 4,274) Tertile 3 (n= 2,575) p-value
Sociodemographic and family factors
Age (range: 11–21), median (Q1–Q3) 15 (14, 17) 15 (14, 16) 15 (14, 16) <0.001
Female, n(%) 2,535 (56.24) 2,179 (50.98) 1,366 (53.05) <0.001
Race/ethnicity, n(%) <0.001
White 2,011 (48.10) 2,397 (56.12) 1,551 (60.26)
Black 852 (20.38) 839 (19.64) 559 (21.72)
Hispanic 826 (19.76) 654 (15.31) 264 (10.26)
Asian 342 (8.18) 225 (5.27) 107 (4.16)
Other 150 (3.59) 156 (3.65) 93 (3.61)
Born in the U.S., n(%) 3,799 (90.84) 4,003 (93.68) 2,435 (94.56) <0.001
Geographic region, n(%) <0.001
Northeast 1,071 (25.60) 1,008 (23.58) 466 (18.10)
Midwest 995 (23.78) 1,140 (26.67) 765 (29.71)
South 1,563 (37.36) 1,567 (36.66) 1,011 (39.26)
West 555 (13.26) 559 (13.08) 333 (12.93)
Two-parent household, n(%) 2,700 (64.58) 2,969 (69.50) 1,865 (72.48) <0.001
Number of siblings (range: 0–12), median (Q1–Q3) 1 (1,2) 1 (1,2) 1 (1,2) <0.001
Household income, n(%) <0.001
1st quintile 783 (24.88) 592 (17.46) 307 (14.87)
2nd quintile 701 (22.28) 670 (19.76) 354 (17.15)
3rd quintile 638 (20.27) 715 (21.09) 422 (20.45)
4th quintile 520 (16.52) 670 (19.76) 457 (22.14)
5th quintile 505 (16.05) 743 (21.92) 524 (25.39)
Household welfare receipt, n(%) 956 (27.07) 731 (19.27) 386 (16.86) <0.001
Has health insurance, n(%) 3,131 (86.02) 3,436 (89.04) 2,127 (90.82) <0.001
Smoker in household, n(%) 1,745 (48.54) 1,679 (43.77) 952 (41.02) <0.001
Mother age (range: 23–81), median (Q1–Q3) 40 (37,45) 41 (37,45) 41 (38,45) 0.005
Mother race/ethnicity, n(%) <0.001
White 1,757 (54.60) 2,215 (63.03) 1,418 (66.35)
Black 596 (18.52) 611 (17.39) 417 (19.51)
Hispanic 604 (18.77) 454 (12.92) 188 (8.80)
Asian 188 (5.84) 152 (4.33) 61 (2.85)
Other 73 (2.27) 82 (2.33) 53 (2.48)
Parents born in the U.S., n(%) 2,869 (72.47) 3,329 (80.57) 2,097 (83.98) <0.001
Parental education, n(%) <0.001
Less than high school 627 (15.11) 411 (9.67) 156 (6.07)
High school equivalency 1,261 (30.39) 1,071 (25.19) 534 (20.76)
Some college 1,038 (25.01) 1,147 (26.98) 686 (26.67)
College degree or higher 1,224 (29.49) 1,623 (38.17) 1,196 (46.50)
Mother employed full-time, n(%) 2,336 (56.28) 2,492 (58.65) 1,547 (60.52) 0.002
Mother religious service attendance, n(%) 0.001
Never or seldom 653 (20.09) 641 (18.17) 353 (16.41)
Less than once a week 1,393 (42.86) 1,516 (42.98) 894 (41.56)
At least once a week 1,204 (37.05) 1,370 (38.84) 904 (42.03)
Mother self-rated health (range: 1–5), median (Q1–Q3) 3 (3, 4) 4 (3, 4) 4 (3, 5) <0.001
Mother happy, n(%) 3,054 (85.52) 3,396 (88.55) 2,104 (91.08) <0.001
(Continued)
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 8 / 17

Table 1. (Continued)
Positive affect
Characteristic Tertile 1 (n= 4,184) Tertile 2 (n= 4,274) Tertile 3 (n= 2,575) p-value
Parent has a disability, n(%) 631 (15.88) 491 (11.87) 306 (12.26) <0.001
Parent has obesity, n(%) 850 (24.43) 859 (23.22) 539 (23.92) 0.482
Parent has alcoholism, n(%) 582 (17.59) 599 (16.84) 301 (13.92) 0.001
Childhood maltreatment by parents, n(%) 979 (23.77) 836 (19.76) 445 (17.43) <0.001
Psychosocial and academic factors
Mental health condition diagnosis
a,b
,n(%) 251 (6.00) 197 (4.61) 109 (4.23) 0.001
Negative affect
b
(range: 0–3), median (Q1–Q3) 1.71 (1.29, 2.00) 1.43 (1.14, 1.71) 1.29 (1.14, 1.57) <0.001
Self-esteem (range: 1–5), median (Q1–Q3) 4 (3.50, 4.33) 4.17 (3.83, 4.50) 4.5 (4.00, 4.83) <0.001
Life expectancy (range: 1–5), median (Q1–Q3) 4.5 (4.00, 5.00) 4.50 (4.00, 5.00) 4.50 (4.50, 5.00) <0.001
Parental control (range: 0–7), median (Q1–Q3) 2 (1, 3) 2 (1, 3) 2 (1, 3) <0.001
Neighborhood social cohesion (range: 0–5), median (Q1–Q3) 4 (3, 5) 4 (3, 5) 4 (3, 5) <0.001
Relationship quality with a parent
a
(range: 1–5), median (Q1–Q3) 5 (4, 5) 5 (4, 5) 5 (5, 5) <0.001
Religious service attendance, n(%) <0.001
Never or seldom 1,132 (27.09) 987 (23.14) 503 (19.54)
Less than once a week 1,510 (36.13) 1,596 (37.41) 901 (35.00)
At least once a week 1,537 (36.78) 1,683 (39.45) 1,170 (45.45)
Has romantic partner, n(%) 1,441 (34.82) 1,478 (34.89) 844 (33.15) 0.283
Has a learning disability, n(%) 564 (15.72) 415 (10.87) 185 (7.97) <0.001
PPVT (range: 13–146), median (Q1–Q3) 96 (87,108) 102 (93,112) 105 (96,115) <0.001
School connectedness (range: 1–5), median (Q1–Q3) 3.60 (3.00, 4.00) 3.80 (3.40, 4.20) 4.00 (3.60, 4.40) <0.001
GPA (range: 1–4), median (Q1–Q3) 2.67 (2.00, 3.25) 2.75 (2.25, 3.50) 3.00 (2.50, 3.67) <0.001
Delinquency (range: 0–15), median (Q1–Q3) 3 (1, 5) 2 (1, 4) 1 (0, 3) <0.001
Health status and health behavior
Somatic symptoms (range: 0–4), median (Q1–Q3) 1.82 (1.55, 2.09) 1.73 (1.45, 2) 1.64 (1.45, 1.91) <0.001
Pubertal development (range: −10.23–9.59), median (Q1–Q3) −0.03 (−1.54, 1.29) 0.23 (−1.14, 1.42) 0.36 (−0.83, 1.68) <0.001
Physical health condition diagnosis
a,b
,n(%) 822 (23.12) 867 (22.88) 512 (22.16) 0.686
Overweight/obesity
b
,n(%) 940 (23.30) 911 (21.82) 479 (18.94) <0.001
Functional limitations
b
,n(%) 33 (0.80) 19 (0.45) 13 (0.51) 0.091
Self-rated health
b
(range: 1–5), median (Q1–Q3) 4 (3, 4) 4 (3, 5) 4 (4, 5) <0.001
Suicidal ideation
b
,n(%) 800 (19.32) 485 (11.43) 201 (7.85) <0.001
Sleep disturbance
b
,n(%) 1,258 (30.08) 988 (23.12) 452 (17.56) <0.001
Physical inactivity
b
,n(%) 259 (6.19) 189 (4.42) 96 (3.73) <0.001
Cigarette smoking
b
,n(%) 802 (19.33) 632 (14.87) 281 (10.97) <0.001
Binge drinking
b
,n(%) 294 (7.04) 223 (5.23) 103 (4.00) <0.001
Marijuana use
b
,n(%) 751 (17.95) 580 (13.57) 266 (10.33) <0.001
Illicit drug use
b
,n(%) 546 (13.32) 482 (11.43) 186 (7.31) <0.001
History of STIs
a
,n(%) 116 (2.78) 89 (2.08) 41 (1.59) 0.004
Preventative health care use
a
,n(%) 2,561 (61.43) 2,850 (66.79) 1,820 (70.82) <0.001
Table is based on non-imputed data. Range for each tertile of positive affect: tertile 1: 1.00–2.75; tertile 2: 3.00–3.50; tertile 3: 3.67–4.00. The statistics presented are
unweighted and sample is restricted to participants surveyed at the exposure wave (Wave II) and first outcome wave (Wave IV), and who had complete data on the
exposure (N= 11,033). The difference in sample size between Table 1 and the outcomes from Table 2 that apply the same sample restrictions (i.e., participation in Wave
II and Wave IV) relates to the 7 participants who had missing data on the exposure measure and had those data imputed for the main analysis, resulting in an overall
sample size for Wave IV outcomes of N= 11,040. No proportion varied by more than +/−4% when restricting the sample by Wave V participation instead of Wave IV
participation. See Table A in S1 Appendix for missing data description, and p-values in this table come from χ2 or analysis of variance tests. Cumulative percentages for
categorical variables may not add up to 100% due to rounding.
a
Pre-baseline covariate for Wave IV outcome.
b
Pre-baseline covariate for Wave V outcome.
PPVT, Peabody Picture Vocabulary Test; STI, sexually transmitted infection.
https://doi.org/10.1371/journal.pmed.1004365.t001
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 9 / 17

Table 2. Associations of positive affect in adolescence with subsequent health and well-being in adulthood (National Longitudinal Study of Adolescent to Adult
Health [Add Health]).
Positive affect
Tertile 1
a
Tertile 2
b
Tertile 3
c
Outcome (Reference) β[95% CI] RR/OR [95% CI] p-value β[95% CI] RR/OR [95% CI] p-value
Physical health
Number of diagnosed physical health
conditions
0.00 −0.03 [−0.09, 0.04] - 0.431 −0.04 [−0.11, 0.03] - 0.239
Cancer 1.00 - 1.24 [0.75, 2.03] 0.403 - 1.02 [0.58, 1.77] 0.957
High cholesterol 1.00 - 0.94 [0.75, 1.18] 0.610 - 0.91 [0.72, 1.15] 0.438
Hypertension 1.00 - 0.89 [0.77, 1.02] 0.087 - 0.90 [0.79, 1.02] 0.102
Diabetes 1.00 - 1.24 [0.93, 1.66] 0.134 - 1.16 [0.79, 1.70] 0.457
Asthma 1.00 - 0.99 [0.85, 1.15] 0.859 - 0.95 [0.79, 1.14] 0.558
Sleep apnea 1.00 - 1.02 [0.85, 1.22] 0.813 - 0.97 [0.78, 1.21] 0.767
Migraines
d
1.00 - 0.84 [0.72, 0.99] 0.034 - 0.79 [0.67, 0.93] 0.005
Allostatic load 0.00 −0.04 [−0.12, 0.04] - 0.266 −0.06 [−0.14, 0.02] - 0.149
Overweight/obesity 1.00 - 0.99 [0.91, 1.09] 0.864 - 0.99 [0.91, 1.07] 0.802
Functional limitations 1.00 - 1.08 [0.90, 1.30] 0.412 - 0.84 [0.69, 1.01] 0.066
Cognition
d
0.00 0.09 [−0.01, 0.17] - 0.019 0.12 [0.05, 0.19] - 0.002
Self-rated health 0.00 −0.00 [−0.06, 0.06] - 0.916 0.11 [0.05, 0.18] - <0.001
Health behavior
Sleep disturbance 1.00 - 0.96 [0.90, 1.02] 0.172 - 0.91 [0.85, 0.97] 0.004
Physical inactivity 1.00 - 0.82 [0.68, 1.00] 0.005 - 0.80 [0.66, 0.98] 0.029
Cigarette smoking 1.00 - 0.92 [0.83, 1.03] 0.141 - 0.93 [0.83, 1.05] 0.258
Binge drinking 1.00 - 0.74 [0.61, 0.91] 0.005 - 0.85 [0.68, 1.05] 0.123
Marijuana use 1.00 - 1.00 [0.88, 1.13] 0.966 - 1.00 [0.87, 1.15] 0.962
Prescription drug misuse 1.00 - 0.93 [0.76, 1.14] 0.481 - 0.72 [0.56, 0.93] 0.013
Illicit drug use 1.00 - 0.92 [0.64, 1.32] 0.638 - 0.92 [0.61, 1.38] 0.674
History of STIs
d
1.00 - 1.15 [0.97, 1.36] 0.104 - 1.10 [0.92, 1.31] 0.307
Preventative health care use
d
1.00 - 0.98 [0.93, 1.03] 0.448 - 0.96 [0.92, 1.01] 0.158
Mental health
Depression diagnosis 1.00 - 0.98 [0.87, 1.10] 0.737 - 0.84 [0.74, 0.94] 0.004
Anxiety diagnosis 1.00 - 0.88 [0.78, 0.99] 0.038 - 0.81 [0.71, 0.93] 0.003
PTSD diagnosis 1.00 - 0.86 [0.65, 1.15] 0.313 - 0.63 [0.46, 0.85] 0.003
ADD/ADHD diagnosis
d
1.00 - 0.87 [0.60, 1.27] 0.472 - 0.66 [0.45, 0.98] 0.039
Negative affect 0.00 −0.05 [−0.12, 0.01] - 0.113 −0.17 [−0.24,
−0.10]
-<0.001
Suicidal ideation 1.00 - 0.74 [0.54, 1.00] 0.052 - 0.74 [0.52, 1.05] 0.089
Perceived stress 0.00 −0.12 [−0.18,
−0.05]
-<0.001 −0.23 [−0.30,
−0.16]
-<0.001
Psychological well-being
Optimism 0.00 0.07 [−0.00, 0.14] - 0.060 0.20 [0.12, 0.28] - <0.001
Job satisfaction
d
0.00 0.08 [0.02, 0.14] - 0.015 0.06 [−0.01, 0.13] - 0.081
Sense of control
d
0.00 0.08 [0.02, 0.13] - 0.009 0.18 [0.13, 0.24] - <0.001
Social factors
Relationship quality with parent 0.00 0.06 [−0.02, 0.14] - 0.168 0.04 [−0.03, 0.12] - 0.279
Social activities 1.00 - 1.00 [0.92, 1.08] 0.973 - 1.06 [0.98, 1.14] 0.142
Social support 0.00 0.06 [0.01, 0.11] - 0.024 0.08 [0.02, 0.13] - 0.008
Loneliness
d
0.00 −0.02 [−0.08, 0.05] - 0.575 −0.09 [−0.16,
−0.02]
- 0.015
(Continued)
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 10 / 17

drug misuse (RR = 072, 95% CI = 056, 093, p= 0.013), lower likelihood of physical inactivity
(RR = 080, 95% CI = 066, 098, p= 0.029), and lower likelihood of sleep disturbance (RR =
091, 95% CI = 085, 097, p= 0.004). However, there was little evidence of an association with
cigarette smoking, binge drinking, marijuana use, illicit drug use, history of STIs, and preven-
tative healthcare use.
When considering psychological outcomes, positive affect was associated with healthier
functioning, or scores, on 6 (out of 7) mental health outcomes, including: reduced likelihood
of PTSD (odds ratio [OR] = 063, 95% CI = 046, 085, p= 0.003), reduced likelihood of ADD/
ADHD (OR = 066, 95% CI = 045, 098, p= 0.039), reduced likelihood of anxiety diagnosis
(RR = 081, 95% CI = 071, 093, p= 0.003), reduced likelihood of depression diagnosis
(RR = 084, 95% CI = 074, 094, p= 0.004), lower perceived stress (β=−023, 95% CI = −030,
−016, p<0.001), and negative affect (β=−017, 95% CI = −024, −010, p<0.001). However,
there was little evidence of an association with suicidal ideation. Positive affect was also
Table 2. (Continued)
Positive affect
Tertile 1
a
Tertile 2
b
Tertile 3
c
Outcome (Reference) β[95% CI] RR/OR [95% CI] p-value β[95% CI] RR/OR [95% CI] p-value
Romantic relationship quality
d
0.00 0.05 [−0.01, 0.11] - 0.109 0.05 [−0.01, 0.12] - 0.121
Satisfaction with parenting
d,e
0.00 0.06 [−0.04, 0.16] - 0.241 0.16 [0.08, 0.25] - <0.001
Perceived discrimination 0.00 −0.04 [−0.11, 0.03] - 0.251 −0.10 [−0.18,
−0.02]
- 0.016
Civic and prosocial behavior
Voting 1.00 - 1.12 [1.04, 1.21] 0.005 - 1.25 [1.16, 1.36] <0.001
Volunteering 1.00 - 1.21 [1.10, 1.33] <0.001 - 1.06 [0.96, 1.18] 0.260
Outcomes were derived from Wave V and models were weighted by the Wave V sample weight unless otherwise noted. Outcomes associated with Wave V: N= 9,003;
outcomes associated with Wave IV: N= 11,040.
The analytic sample was restricted to those who participated in the survey at the exposure wave (Wave II) and had a valid sampling weight at the outcome wave from
which the data for the respective outcome was derived (Wave IV or Wave V). Multiple imputation was performed to impute missing data on the covariates, exposure,
and outcomes. All models controlled for sociodemographic and family factors (age, sex, race/ethnicity, nativity status, geographic region, family structure, number of
siblings, household income, household welfare receipt, insurance status, smoker in household, mother age, mother race/ethnicity, parent nativity, parental education,
mother employment status, mother religious service attendance, mother health status, mother happiness, parent has a disability, parent has obesity, parent has
alcoholism, childhood maltreatment by parents), psychosocial and academic factors (mental health condition diagnosis, negative affect, self-esteem, life orientation,
relationship quality with a parent, parental control, neighborhood social cohesion, religious service attendance, romantic relationship status, has a learning disability,
PPVT, school connectedness, GPA, delinquency), health status and health behavior (somatic symptoms, pubertal development, physical health condition diagnosis,
overweight/obesity, functional limitations, self-rated health, suicidal ideation, sleep disturbance, physical inactivity, cigarette smoking, binge drinking, marijuana use,
illicit drug use, history of STIs, preventative health care use), and positive affect assessed at Wave I.
An outcome-wide analytic approach was used, and a separate model was run for each outcome. A different type of model was run depending on the nature of the
outcome: (1) for each binary outcome with a prevalence of 10%, a generalized linear model (with a log link and Poisson distribution) was used to estimate an RR; (2)
for each binary outcome with a prevalence of <10%, a logistic regression model was used to estimate an OR; and (3) for each continuous outcome, a linear regression
model was used to estimate a β.
All continuous outcomes were standardized (mean = 0, standard deviation = 1), and βwas the standardized effect size.
a
For outcomes associated with Wave V: Tertile 1 n= 3,354; for outcomes associated with Wave IV: Tertile 1 n= 4,186.
b
For outcomes associated with Wave V: Tertile 2 n= 2,422; for outcomes associated with Wave IV: Tertile 2 n= 3,015.
c
For outcomes associated with Wave V: Tertile 3 n= 3,227; for outcomes associated with Wave IV: Tertile 3 n= 3,839.
d
Outcome was derived from data from Wave IV and model was weighted by the Wave IV sample weight because the data for this outcome was not collected at Wave V.
e
Analysis for this outcome was restricted to participants who reported having at least 1 child at Wave IV (n= 5,304).
ADHD, attention deficit hyperactivity disorder; CI, confidence interval; OR, odds ratio; PTSD, posttraumatic stress disorder; RR, risk ratio;.STI, sexually transmitted
infection.
https://doi.org/10.1371/journal.pmed.1004365.t002
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 11 / 17

associated with healthier functioning, or scores, on 2 (out of 3) psychological well-being out-
comes, including: optimism (β= 020, 95% CI = 012, 028, p<0.001) and sense of control (β
= 018, 95% CI = 013, 024, p<0.001), but there was little evidence of association with job
satisfaction.
When considering social outcomes, positive affect was associated healthier functioning, or
scores, on 4 (out of 7) social factors, including: lower perceived discrimination (β=−010, 95%
CI = −018, −002, p= 0.016) and loneliness (β=−009, 95% CI = −016, −002, p= 0.015), as
well as higher satisfaction with parenting (β= 016, 95% CI = 008, 025, p<0.001) and social
support (β= 008, 95% CI = 002, 013, p= 0.008). However, there was little evidence of associ-
ation with relationship quality with parent, social activities, or romantic relationship quality.
Finally, positive affect was associated with 1 (out of 2) civic and prosocial behavior factors
including: increased voting (RR = 125, 95% CI = 116, 136, p<0.001). However, there was lit-
tle evidence of association with volunteering.
We conducted 4 additional analyses. First, E-value analyses suggested that a few of the asso-
ciations we observed were at least moderately robust to unmeasured confounding (Table 3).
For example, an unmeasured confounder associated with both positive affect and anxiety diag-
nosis by risk ratios of 176, each, above, and beyond the large array of potential confounders
already adjusted for, could explain away the association. However, weaker joint confounder
associations could not. To shift the confidence interval to include the null, an unmeasured
confounder associated with both positive affect and anxiety diagnosis by risk ratios of 136
each could suffice, but weaker joint confounder associations could not. However, several other
associations were not especially robust to potential unmeasured confounding. Second, com-
plete-case analyses provided similar results to those in the main analyses (Table C in
S1 Appendix). Third, unadjusted models provided a baseline for comparison (Table D in
S1 Appendix). Fourth, conventionally adjusted covariate models showed estimates that were
stronger than the fully adjusted models (Table E in S1 Appendix). Fifth, we calculated the
NNT and NNH for each outcome (see Table F in S1 Appendix).
Discussion
In a nationally representative sample of U.S. adolescents, we observed that higher positive
affect during adolescence was associated with many health/well-being outcomes in adulthood.
These results were maintained after robust control for a wide range of potential confounders,
as well as positive affect (and all the outcomes when available) in the prior wave. Positive affect
was associated with most mental health outcomes (i.e., lower likelihood of PTSD diagnosis,
ADD/ADHD diagnosis, anxiety diagnosis, and depression diagnosis, along with lower per-
ceived stress and negative affect) and the majority of social outcomes (i.e., lower perceived dis-
crimination and loneliness, as well as higher satisfaction with parenting and social support). It
was also associated with some health behaviors (i.e., lower likelihood of prescription drug mis-
use, physical inactivity, and sleep disturbance) and a few physical health outcomes (i.e., lower
likelihood of migraines, higher cognition, and self-rated health). Finally, positive affect was
associated with most psychological well-being outcomes (i.e., higher optimism and sense of
control), and also a civic/prosocial outcome (i.e., more voting). It is important to note that
even by the end of follow-up, participants were relatively young and many physical health con-
ditions typically emerge later in life.
Our results share some alignment with results from past work that evaluated “prevalence”
of positive affect and outcomes. For example, consistent with past research which focused on
adolescents and adults, we observed that “incident” positive affect was associated with some
higher psychological well-being outcomes (e.g., sense of control and optimism) and better
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 12 / 17

Table 3. Robustness to unmeasured confounding (E-values) for the association between positive affect (3rd tertile
vs. 1st tertile) in adolescence and subsequent health and well-being in adulthood (National Longitudinal Study of
Adolescent to Adult Health [Add Health]).
Outcome Effect estimate
a
Confidence interval limit
b
Physical health
Number of diagnosed physical health conditions 1.23 1.00
Cancer 1.14 1.00
High cholesterol 1.42 1.00
Hypertension 1.48 1.00
Diabetes 1.58 1.00
Asthma 1.29 1.00
Sleep apnea 1.21 1.00
Migraines
c
1.86 1.37
Allostatic load 1.29 1.00
Overweight/obesity 1.11 1.00
Functional limitations 1.68 1.00
Cognition
c
1.47 1.26
Self-rated health 1.46 1.27
Health behavior
Sleep disturbance 1.44 1.21
Physical inactivity 1.80 1.18
Cigarette smoking 1.35 1.00
Binge drinking 1.65 1.00
Marijuana use 1.06 1.00
Prescription drug misuse 2.13 1.36
Illicit drug use 1.41 1.00
History of STIs
c
1.42 1.00
Preventative health care use
c
1.23 1.00
Mental health
Depression diagnosis 1.68 1.31
Anxiety diagnosis 1.76 1.36
PTSD diagnosis 2.58 1.64
ADD/ADHD diagnosis
c
2.38 1.17
Negative affect 1.61 1.42
Suicidal ideation 2.06 1.00
Perceived stress 1.76 1.58
Psychological well-being
Optimism 1.69 1.48
Job satisfaction
c
1.31 1.00
Sense of control
c
1.65 1.50
Social factors
Relationship quality with parent 1.28 1.00
Social activities 1.31 1.00
Social support 1.35 1.16
Loneliness
c
1.39 1.15
Romantic relationship quality
c
1.28 1.00
Satisfaction with parenting
c
1.59 1.37
Perceived discrimination 1.42 1.16
Civic and prosocial behavior
Voting 1.82 1.58
(Continued)
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 13 / 17

mental health outcomes (e.g., lower likelihood of/scores: negative affect, depression diagnosis,
anxiety diagnosis, perceived stress) [33]. Also aligned with some past research we observed
null associations (e.g., no associations with: smoking) [14]. However, our results also diverge
with results from past research. For example, contrary to prior research which observed associ-
ations with some health behaviors [14,17,18], we did not observe associations with some health
behaviors (e.g., not: smoking, binge drinking). We also did not observe associations with some
physical health outcomes (e.g., healthier: BMI and composite biomarker scores) that past stud-
ies observed [14–16]. However, when considering our results that adjusted for only sociode-
mographic covariates we observed associations that align with prior research, such as
associations with some health behaviors (i.e., lower likelihood of: smoking, binge drinking)
and some health outcomes (i.e., composite biomarker score). Methodologically, the underlying
reasons for diverging results between our study and past studies may stem from a range of
sources including differences in: (1) which covariates were controlled for; (2) control for prior
positive affect; (3) study population (e.g., nationally representative versus non-generalizable
samples); (4) study design (e.g., cross-sectional versus longitudinal); (5) measurement of the
exposure; and (6) measurement of the outcome (e.g., specific versus composite measures).
Our study has several limitations. First, both positive affect and most outcomes were self-
reported; thus, potential self-report and common method bias is a concern. However, control
for pre-baseline outcomes and a wide range of potential confounders helps to mitigate these
concerns. Second, confounding by unmeasured variables and reverse causality are common
concerns in observational research. However, controlling for a large array of variables, includ-
ing the exposure and outcomes in the pre-baseline wave, the prospective nature of our data,
and results from E-value analyses helps mitigate these concerns. Third, positive affect was
derived from a scale originally designed to assess depressive symptoms. However, this subscale
has demonstrated reliability and validity in various studies, and the positive affect dimension
repeatedly emerges in factor analytics studies as illustrated by a meta-analysis [24], and has
been used repeatedly in past research [25]. Future studies should consider using other positive
affect scales (e.g., PANAS items, facial coding). Fourth, our study focused on U.S. adolescents,
and its findings may not extend to other cultural contexts, where different cultural, social, and
environmental factors can influence health and development, potentially affecting the applica-
bility of our results internationally. Our study also has several important strengths including
the use of a prospective, diverse, and nationally representative sample of adolescents. The
study allowed us to evaluate evidence for a distinct question often of more interest to policy-
Table 3. (Continued)
Outcome Effect estimate
a
Confidence interval limit
b
Volunteering 1.32 1.00
The formula for calculating E-values can be found in VanderWeele and Ding (2017)
Outcomes were derived from Wave V unless otherwise noted.
a
E-values for effect estimates are the minimum strength of association on the risk ratio scale that an unmeasured
confounder would need to have with both the exposure and the outcome to fully explain away the observed
association between the exposure and outcome, conditional on the measured covariates.
b
E-values for the limit of the 95% CI closest to the null denote the minimum strength of association on the risk ratio
scale that an unmeasured confounder would need to have with both the exposure and the outcome to shift the CI to
include the null value, conditional on the measured covariates.
c
Outcome was derived from data from Wave IV because the data for this outcome was not collected at Wave V.
ADHD, attention deficit hyperactivity disorder; CI, confidence interval; PTSD, posttraumatic stress disorder; STI,
sexually transmitted infection.
https://doi.org/10.1371/journal.pmed.1004365.t003
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 14 / 17

makers and interventionists. However, only 1 year of positive affect exposure, which is what
the design here evaluates, may be insufficiently long to substantially affect outcomes 10 to 12
or 19 to 22 years later. Evaluating the effects of longer positive affect exposure periods may be
important.
An increasing number of countries are adopting well-being measures as critical tools for
guiding policy choices [1]. Further, several countries seek innovative and cost-effective meth-
ods of enhancing the health/well-being of large adolescent populations. Evidence from both
randomized controlled trials (aimed at individuals) [11] and case studies of successful policies
(aimed at entire populations) [12] suggest positive affect can be enhanced. Our findings sug-
gest that ongoing development and application of interventions and policies aimed at bolster-
ing positive affect is a promising method of enhancing some aspects of health/well-being for
our adolescent and emerging adult populations.
Supporting information
S1 Appendix. Text A. Assessment of Outcomes. Text B. Proof Illustrating How Controlling
for Prior Levels of Positive Affect Can Help Us Evaluate How “Change” in Positive Affect is
Associated with Subsequent Health and Well-Being Outcomes Over Time. Table A. Missing
Data on Study Variables (National Longitudinal Study of Adolescent to Adult Health [Add
Health]). Table B. Change in Positive Affect from the Pre-Baseline Wave (Wave I; t
0
) to the
Baseline Wave (Wave II; t
1
). Table C. Associations of Positive Affect in Adolescence with Sub-
sequent Health and Well-Being in Adulthood (Complete-Case Analyses; National Longitudi-
nal Study of Adolescent to Adult Health [Add Health]). Table D. Associations of Positive
Affect in Adolescence with Subsequent Health and Well-Being in Adulthood (Unadjusted or
Fully Adjusted for Covariates; National Longitudinal Study of Adolescent to Adult Health
[Add Health]). Table E. Associations of Positive Affect in Adolescence with Subsequent
Health and Well-Being in Adulthood (Adjusting for Conventional Covariates or All Covari-
ates; National Longitudinal Study of Adolescent to Adult Health [Add Health]). Table F. Asso-
ciations of Positive Affect in Adolescence with Subsequent Health and Well-Being in
Adulthood (Actual Amounts and/or Absolute Risks of Binary Outcomes; National Longitudi-
nal Study of Adolescent to Adult Health [Add Health]). Fig A. Sample Inclusion Criteria for
Positive Affect Analyses (Wave IV Outcomes). Fig B. Sample Inclusion Criteria for Positive
Affect Analyses (Wave V Outcomes). Checklist A. Strengthening the Reporting of Observa-
tional Studies in Epidemiology (STROBE) Checklist.
(DOCX)
Acknowledgments
This research uses data from Add Health, funded by grant P01 HD31921 (Harris) from the
Eunice Kennedy Shriver National Institute of Child Health and Human Development
(NICHD), with cooperative funding from 23 other federal agencies and foundations. Add
Health is currently directed by Robert A. Hummer and funded by the National Institute on
Aging cooperative agreements U01 AG071448 (Hummer) and U01AG071450 (Aiello and
Hummer) at the University of North Carolina at Chapel Hill. Add Health was designed by J.
Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris at the University of North Caro-
lina at Chapel Hill.
Author Contributions
Conceptualization: Eric S. Kim, Renae Wilkinson, Tyler J. VanderWeele.
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 15 / 17

Formal analysis: Renae Wilkinson.
Funding acquisition: Eric S. Kim, Tyler J. VanderWeele.
Supervision: Eric S. Kim, Tyler J. VanderWeele.
Writing – original draft: Eric S. Kim.
Writing – review & editing: Eric S. Kim, Renae Wilkinson, Sakurako S. Okuzono, Ying Chen,
Koichiro Shiba, Richard G. Cowden, Tyler J. VanderWeele.
References
1. Stiglitz J E, Fitoussi J-P, Durand M. Beyond GDP Measuring What Counts for Economic and Social Per-
formance: Measuring What Counts for Economic and Social Performance. OECD Publishing; 2018.
2. Patton G C, Sawyer S M, Santelli J S, Ross D A, Afifi R, Allen N B, et al. Our future: a Lancet commis-
sion on adolescent health and wellbeing. Lancet. 2016; 387:2423–2478. https://doi.org/10.1016/S0140-
6736(16)00579-1 PMID: 27174304
3. Kubzansky L D, Huffman J C, Boehm J K, Hernandez R, Kim E S, Koga H K, et al. Positive psychologi-
cal well-being and cardiovascular health promotion. J Am Coll Cardiol. 2018; 72:1382–1396. https://doi.
org/10.1016/j.jacc.2018.07.042 PMID: 30213332
4. VanderWeele T J. On the promotion of human flourishing. Proc Natl Acad Sci U S A. 2017; 114:8148–
8156. https://doi.org/10.1073/pnas.1702996114 PMID: 28705870
5. Ryff C D. Psychological well-being revisited: advances in the science and practice of eudaimonia. Psy-
chother Psychosom. 2014; 83:10–28. https://doi.org/10.1159/000353263 PMID: 24281296
6. Seligman M E P. Positive health. Appl Psychol Int Rev. 2008; 57:3–18. https://doi.org/10.1111/j.1464-
0597.2008.00351.x
7. Steptoe A. Happiness and health. Annu Rev Public Health. 2019; 40:339–359. https://doi.org/10.1146/
annurev-publhealth-040218-044150 PMID: 30601719
8. Pressman S D, Jenkins BN, Moskowitz JT. Positive affect and health: what do we know and where next
should we go? Annu Rev Psychol. 2019; 70:627–650. https://doi.org/10.1146/annurev-psych-010418-
102955 PMID: 30260746
9. Twenge J M, Martin G N, Campbell W K. Decreases in psychological well-being among American ado-
lescents after 2012 and links to screen time during the rise of smartphone technology. Emot. 2018;
18:765. https://doi.org/10.1037/emo0000403 PMID: 29355336
10. Diener E, Oishi S, Tay L. Advances in subjective well-being research. Nat Hum Behav. 2018; 2:253–
260. https://doi.org/10.1038/s41562-018-0307-6 PMID: 30936533
11. van Agteren J, Iasiello M, Lo L, Bartholomaeus J, Kopsaftis Z, Carey M, et al. A systematic review and
meta-analysis of psychological interventions to improve mental wellbeing. Nat Hum Behav. 2021;
5:631–652. https://doi.org/10.1038/s41562-021-01093-w PMID: 33875837
12. The Global Council for Happiness and Wellbeing. Global Happiness and Wellbeing Policy Report. New
York: Sustainable Development Solutions Network; 2019.
13. Martı
´n-Marı
´a N, Miret M, Caballero F F, Rico-Uribe L A, Steptoe A, Chatterji S, et al. The impact of sub-
jective well-being on mortality: a meta-analysis of longitudinal studies in the general population. Psycho-
som Med. 2017; 79:565–575. https://doi.org/10.1097/PSY.0000000000000444 PMID: 28033196
14. Boehm J K, Qureshi F, Kubzansky L D. Child psychological well-being and adult health behavior and
body mass index. Health Psychol. 2023; 42:73. https://doi.org/10.1037/hea0001261 PMID: 36595459
15. Qureshi F, Guimond A, Tsao E, Delaney S, Boehm J K, Kubzansky L D. Adolescent psychological
assets and cardiometabolic health maintenance in adulthood: implications for health equity. J Am Heart
Assoc. 2023:e026173. https://doi.org/10.1161/JAHA.122.026173 PMID: 36628968
16. Jeffery A N, Hyland M E, Hosking J, Wilkin T J. Mood and its association with metabolic health in adoles-
cents: a longitudinal study, EarlyBird 65. Pediatr Diabetes. 2014; 15:599–605. https://doi.org/10.1111/
pedi.12125 PMID: 24552539
17. Hoyt L T, Chase-Lansdale P L, McDade T W, Adam E K. Positive youth, healthy adults: does positive
well-being in adolescence predict better perceived health and fewer risky health behaviors in young
adulthood? J Adolesc Health. 2012; 50:66–73. https://doi.org/10.1016/j.jadohealth.2011.05.002 PMID:
22188836
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 16 / 17

18. Mantey D S, Clendennen S L, Sumbe A, Wilkinson A V, Harrell M B. Positive affect and multiple-
tobacco product use among youth: A 3-year longitudinal study. Am J Health Behav. 2021; 45:849–855.
https://doi.org/10.5993/AJHB.45.5.5 PMID: 34702432
19. Kalak N, Lemola S, Brand S, Holsboer–Trachsler E, Grob A. Sleep duration and subjective psychologi-
cal well-being in adolescence: a longitudinal study in Switzerland and Norway. Neuropsychiatr Dis
Treat. 2014;1199–1207. https://doi.org/10.2147/NDT.S62533 PMID: 25061305
20. VanderWeele T J, Mathur M B, Chen Y. Outcome-wide longitudinal designs for causal inference: a new
template for empirical studies. Stat Sci. 2020; 35:437–466.
21. Duncan P M, Shaw J S, Hagan J F. Bright futures: guidelines for health supervision of infants, children,
and adolescents. Pediatrics. 2008.
22. Sawyer S M, Azzopardi P S, Wickremarathne D, Patton G C. The age of adolescence. Lancet Child
Adolesc Health. 2018; 2:223–228. https://doi.org/10.1016/S2352-4642(18)30022-1 PMID: 30169257
23. Radloff L S. The CES-D Scale: A self-report depression scale for research in the general population.
Appl Psychol Meas. 1977; 1:385–401. https://doi.org/10.1177/014662167700100306
24. Shafer A B. Meta-analysis of the factor structures of four depression questionnaires: Beck, CES-D,
Hamilton, and Zung. J Clin Psychol. 2006; 62:123–146. https://doi.org/10.1002/jclp.20213 PMID:
16287149
25. Boehm J K, Chen Y, Qureshi F, Soo J, Umukoro P, Hernandez R, et al. Positive emotions and favorable
cardiovascular health: A 20-year longitudinal study. Prev Med. 2020; 136:106103. https://doi.org/10.
1016/j.ypmed.2020.106103 PMID: 32348855
26. Steinberg L. Cognitive and affective development in adolescence. Trends Cogn Sci. 2005; 9:69–74.
https://doi.org/10.1016/j.tics.2004.12.005 PMID: 15668099
27. Lerner J V, Phelps E, Forman Y, Bowers E P. Positive youth development. John Wiley & Sons Inc;
2009.
28. Damon W. What is positive youth development? Ann Am Acad Pol Soc Sci. 2004; 591:13–24. https://
doi.org/doi.org/10.1177/0002716203260
29. Greenland S, Robins J M. Identifiability, exchangeability, and epidemiological confounding. Int J Epide-
miol. 1986; 15:413–419. https://doi.org/10.1093/ije/15.3.413 PMID: 3771081
30. Robins J. Estimation of the time-dependent accelerated failure time model in the presence of confound-
ing factors. Biometrika. 1992; 79:321–334. https://doi.org/10.2307/2336843
31. VanderWeele T J, Ding P. Sensitivity analysis in observational research: introducing the E-Value. Ann
Intern Med. 2017; 167:268–274. https://doi.org/10.7326/M16-2607 PMID: 28693043
32. Sterne J A C, White I R, Carlin J B, Spratt M, Royston P, Kenward M G, et al. Multiple imputation for
missing data in epidemiological and clinical research: potential and pitfalls. BMJ. 2009; 338:b2393.
https://doi.org/10.1136/bmj.b2393 PMID: 19564179
33. Kansky J, Allen J P, Diener E. Early adolescent affect predicts later life outcomes. Appl Psychol Health
Well-Being. 2016; 8:192–212. https://doi.org/10.1111/aphw.12068 PMID: 27075545
PLOS MEDICINE
Positive affect outcome-wide
PLOS Medicine | https://doi.org/10.1371/journal.pmed.1004365 April 2, 2024 17 / 17