Content uploaded by Chris Sakellariou
Author content
All content in this area was uploaded by Chris Sakellariou on Apr 25, 2023
Content may be subject to copyright.
Vol.:(0123456789)
Child Indicators Research
https://doi.org/10.1007/s12187-023-10029-x
1 3
The Effect ofBody Image Perceptions onLife Satisfaction
andEmotional Wellbeing ofAdolescent Students:
ChrisSakellariou1
Accepted: 20 March 2023
© The Author(s), under exclusive licence to Springer Nature B.V. 2023
Abstract
Studies investigating specific determinants of subjective wellbeing (such as body
image perceptions) using experimental/quasi-experimental methods are lacking.
Furthermore, few studies considered more than one dimension of wellbeing, used
multi-country samples, or considered a variety of determinants/correlates of well-
being. Only a small minority of studies are on adolescents. I used a large multi-
country sample of 15-year-old students, to implement an innovative methodologi-
cal approach which accounts for potential endogeneity of body image perceptions
and derive estimates of the effect of body image on the cognitive and emotional
wellbeing of adolescent students. I supplemented Instrumental Variables (IV) esti-
mation with the newly developed—instrument free estimation method, to derive
gender-specific causal effect estimates. The outcome measures considered are Life
Satisfaction (0–10 scale) and Positive Affect. I found that biases associated with
endogeneity of perceived body image are more important when emotional wellbeing
is considered. Similarly, gender differences in the effect of body image satisfaction
were established only on the emotional dimension of wellbeing (Positive Affect);
the effect size for girls is about three times larger than for boys.
Keywords Subjective wellbeing· Body image· Quasi-experimental methods·
Gender
1 Introduction
Indicators of overall adult subjective wellbeing (SWB thereafter), such as Life Satisfac-
tion (LS hereafter), have been increasingly used as a measure of adolescent wellbe-
ing. Individual differences in reported wellbeing of adolescents have been associated
with several intrapersonal and interpersonal variables, involving a complex interplay of
* Chris Sakellariou
acsake@ntu.edu.sg
1 School ofSocial Sciences, Nanyang Technological University, 48 Nanyang Avenue, HSS-04-64,
Singapore639818, Singapore
C.Sakellariou
1 3
environmental and personal factors, which include interpersonal relationships, cogni-
tive attributions, self-perceptions, and structured activities (e.g., Huebner etal., 2006).
Life satisfaction studies for children and adolescents followed such studies for
adults, after the appearance of well-validated instruments appropriate for children.
Huebner (1994) proposed and validated the Multidimensional Students’ LS Scale,
which assesses general LS in five specific domains: Family, Friends, School, Self, and
Living Environment (see also, Huebner etal., 1998). The “Self” dimension includes
physical wellbeing, academic self-esteem (i.e., how a person regards himself/herself as
a student), and perceptions of appearance (body image perceptions). Empirical stud-
ies on adolescent students report that, school related factors such as relationships with
school peers and sense of belonging are important (e.g., Lee & Yoo, 2015; Lemma
et al., 2015). On the other hand, demographic, socioeconomic, and related variables
can only explain a small part of individual differences in reported SWB (e.g., Gilman
& Huebner, 2006).
1.1 Body Image Perceptions andSubjective Wellbeing
Evidence from several studies on adults, mostly over the last decade or so, indicates
that body image perceptions and SWB are interlinked, with LS and body dissatisfaction
inversely associated, and LS and body appreciation directly associated (e.g., Swami
etal., 2017). Research on the association between body image perceptions and wellbe-
ing in adolescents is scant. However, being satisfied with your body image is impor-
tant during adolescence. During puberty, adolescents undergo physical changes, which
influence their psychological and social adjustment and health behaviors. Biological,
psychological, and sociocultural factors have been found to be relevant in understand-
ing the development of body image concerns and weight loss strategies among chil-
dren. Sociocultural factors such as media, peers and family are also important for how
adolescents perceive their bodies (for an overview, see Holmqvist Gattario etal., 2014).
While body image dissatisfaction affects both boys and girls, girls are more likely to
perceive that they are overweight, indicating on average a higher level of body dissatis-
faction than adolescent boys (e.g., Yates etal., 2004).
Existing research on the association between body image satisfaction and wellbe-
ing in adolescents mostly focuses on the link between body dissatisfaction and mental
health related outcomes, notably depression, suicidal behavior, emotional wellbeing,
eating disorders, and risky health behaviors. Based on selected empirical literature,
body dissatisfaction in adolescence predicts the occurrence of later depressive episodes
in adolescents (and likely more so for girls), is linked to impairment of emotional well-
being of overweight adolescents, and predicts the occurrence of several risky health
behaviors, such as drug use, self-harm, gambling, and drinking (e.g., Bearman & Stice,
2008; Bornioli etal., 2020; Mond etal., 2011).
1.2 Objectives andContribution ofCurrent Study
From the review of the literature, the objectives of earlier empirical research—
whether on adults or adolescents—were to establish associations between
1 3
The Effect ofBody Image Perceptions onLife Satisfaction and…
hypothesized determinants of SWB (including body image perceptions, as well as
other psychological traits), to identify predictors in the context of the objectives of
research undertaken. Methodological approaches for estimating strength of associa-
tions between predictor and outcome were generally based on multiple regression
analysis, multilevel models /hierarchical linear modeling, or path analysis, with
some using longitudinal data. Such methodologies are suitable for establishing that
body image perceptions indeed influence wellbeing and in addressing the specific
objectives of each of these studies. However, they are less suitable in establishing
the magnitude of causal effects, and the possibility of heterogeneous effects by gen-
der. Findings suggest that body image satisfaction is a significant predictor of LS,
over and above the influence of BMI and personality; however, coefficient estimates
are generally small and of similar magnitude for males and females (e.g., Davies
etal., 2020). The literature is clear that girls tend to have more negative body image
perceptions compared to boys; it would be intuitive to expect gender differences in
effect sizes as well.
Methodologies in existing studies generally account for a subset of observed
differences between study participants (i.e., selection on observables only). How-
ever, self-reported body image should be treated as an endogenous1 covariate,
due to unobserved pre-existing differences between subjects who report higher
body satisfaction and those who report lower body satisfaction. This introduces
bias when estimating effects on the outcome. Given an outcome variable (y)
and a vector of explanatory variables (x), assumes that: (a) vector x is uncorre-
lated with the errors in the equation to be estimated; (b) there is no simultaneity/
reverse causation; and (c) variables in x are measured without error. When we
expect violation of one or more of the above requirements—in isolation or in
combination for one or more of the covariates in x—covariate endogeneity is a
complication. Thinking in terms of the relationship between body image percep-
tions and a wellbeing indicator, correlation between body image and the error
term could arise because we have omitted one or more important covariates from
our model, which are correlated with body image. Alternatively, it could be that
both body image and wellbeing are to some extent determined by the same unob-
served factors. Another potential source of endogeneity, resulting in inconsist-
ent OLS estimates, is that body image perception is self-reported; therefore, the
OLS estimator is likely biased towards zero. Finally, while the dominant effect is
expected to be from body image to subjective wellbeing, one cannot rule out sim-
ultaneity, i.e., body image and subjective wellbeing influence each other simul-
taneously, so that causality is not exclusively from body image to the outcome
variable. Give a valid instrument, IV estimation can deal with simultaneity bias,
along with biases associated with omitted variables and measurement error.
1 By endogenous I refer to a covariate appearing in a model which is correlated with unobserved charac-
teristics which also affect the outcome.
C.Sakellariou
1 3
Two research questions were investigated:
RQ1: Is there evidence of endogeneity of body image perceptions and associated
biases in effect estimates when unobservables are not accounted for? I hypothe-
size that such biases are present and are important when estimating causal effects.
RQ2: Are there gender differences in effect estimates of body image perceptions
on subjective wellbeing? Based on intuition, I expect to find that effect estimates
a larger for girls.
To investigate the above research questions, I modeled perceived body image
as a potentially endogenous covariate when estimating its effect on two indica-
tors of adolescent wellbeing, one on the evaluative dimension (LS), and the
other on the emotional dimension of wellbeing, i.e., frequency of positive emo-
tions (Positive Affect). I followed quasi-experimental approach, by supplement-
ing Instrumental Variables (IV) estimation with the newly developed – instru-
ment free – kinky least squares (KLS) method. Further analysis was conducted
assuming imperfect instruments, using the approach proposed by Nevo and
Rosen (2012). Models were estimated by gender, to un uncover gender differ-
ences in effects. The specifics of the methodological approach are discussed in
the next section.
2 Materials andMethods
2.1 Data
I used data from the 2018 Program for International Student Assessment (PISA)
on 15-year-old students. The compulsory background questionnaires (student
and school questionnaires) contain information on students themselves, as well
as household information, school and school experiences, and students’ atti-
tudes, dispositions, and beliefs. PISA 2018, besides the compulsory background
questionnaires, also included an optional module about students’ wellbeing,
with nine participating countries (Bulgaria, Georgia, Hong Kong, Ireland, Mex-
ico, Panama, Serbia, Spain, and United Arab Emirates) and about 75,000 stu-
dent participants. The wellbeing questionnaire focused on students’ perceptions
about their health, SWB indicators such as LS and SWB-Positive Affect, and
sense of meaning in life (eudemonic wellbeing), attitudes, social relationships,
and activities outside of school. The survey assessed important factors for well-
being of adolescent students, both objective (such as health, living environ-
ments, and financial situation) and subjective (such as individual personality
attributes). Among other wellbeing indicators of students as individuals (i.e.,
health, education, and psychological functioning), perception of, and satisfac-
tion with one’s body image was included in the wellbeing questionnaire.
1 3
The Effect ofBody Image Perceptions onLife Satisfaction and…
2.2 Wellbeing
Wellbeing is a multi-dimensional construct which includes both objective and sub-
jective wellbeing. SWB, a self-reported measure of wellbeing, includes an emo-
tional component (positive and negative affect) and an evaluative/cognitive compo-
nent (overall LS). Psychological factors interact with life circumstances in producing
SWB (e.g., Diener, 1984).
Life satisfaction is an individual’s overall appraisal of her or his quality of life
and has been described as a key indicator of SWB (e.g., Veenhoven, 1988). Global
appraisal of LS by an individual subsumes specific second-order domains, such as
self, family, friends, school, and living environment (e.g., Gilman etal., 2000). In
adolescents, LS is closely related to a variety of life experiences in the family, peer,
and school environments (e.g., Gilman & Huebner, 2003). Approaches in assessing
overall LS involve questions targeting either an evaluative aspect (from “the worst
possible life” to “the best possible life”) or focusing on satisfaction instead of evalu-
ation (“how satisfied are you with your life overall these days” on a scale from 1 to
10), as in PISA 2018.
The SWB approach (Diener, 1984) involves three distinct but related components
of how people perceive their own well-being: Frequency of Positive Affect (expe-
riencing positive emotions and moods); Frequency of negative affect (not experi-
encing negative feelings or moods often); and a Cognitive component (how people
think about their lives and life satisfaction). PISA 2018 contains a SWB-Positive
Affect composite index, based on three survey questions on the frequency 15-year-
olds felt joyful, cheerful, and happy.
2.3 Methodological Approach
Of the empirical approaches used which aim to uncover a causal relationship
while using observational data, some (e.g., multivariate regression, multilevel
modelling, and propensity score matching), do not account for the possibility that
one or more important confounders (observed or unobserved), could be another
common cause of the outcome and the covariate of interest. There are, however,
approaches which allow estimating a quasi-causal effect from data that arise from
non-experimental research designs (see for example, Rutkowski, 2016). These
include quasi-experimental strategies such as the potential outcomes/counter-
factual theory of causality (e.g., Rubin, 1974), use of natural experiments, and
instrumental variables (IV) approaches, which allow considerable control over
omitted variables.
To address complications posed by endogenous covariates, we need to model
potentially endogenous covariates in out model, along with the outcome equation.
One frequently used approach is to follow an Instrumental Variables (IV) approach
which requires using one or more variables (“instruments”) that affect the endog-
enous covariate but can be excluded from the outcome equation.
C.Sakellariou
1 3
2.4 Instrument forPerceived Body Image
The instrument used for body image perceptions in the first stage of IV estimation
is the average body image at the school level, derived using the country-school
identifier. Given that the instrument used is an ordered continuous instrument,
there is a different local average treatment effect (LATE) parameter for every
value of the continuous instrument; hence, the linear instrumental variables esti-
mator estimates a weighted average of local average treatment effects.
Instrument relevance and validity depend on the following assumptions:
Relevance The instrument and the endogenous covariate are sufficiently associated
either because of a causal association between the two, or because the two share a
common cause. In observational studies, it is assumed that there is an unmeasured
causal instrument which is the common cause of the measured proxy instrument and
the endogenous covariate (e.g., Lousdal, 2018). The rationale for instrument rele-
vance is based on the expectation that body image at the school level is positively
correlated with individual students’ body image perceptions, due to students inter-
acting and socializing with one another, thus influencing beliefs and perceptions
such as body image. Instrument relevance can be tested using conventional weak
instrument tests.
Independence The instrument must be uncorrelated with the error term, i.e., the
instrument does not share common causes with the outcome. Independence cannot
be tested, since the error term is, by definition, unobservable.
The Exclusion Restriction The instrument is independent of the outcome, after con-
ditioning for additional covariates, i.e., the instrument affects the outcome only
through its effect on the endogenous covariate. Although the exclusion restriction
is usually considered untestable, I provide evidence supporting the validity of the
exclusion restriction, using a recently available approach which relies on a plausible
range of endogeneity correlations.
The three basic assumptions allow for identification of an upper and lower
bound of the causal effect. Violation of the Exclusion and Independence assump-
tions will result in biased estimates. Violation of the Relevance assumption (weak
instrument), on the other hand, may be associated with IV estimates which are
even more biased than the conventional regression estimates.
An additional assumption needed to point identify a causal effect is that of
Deterministic Monotonicity. This assumption is usually in the context of binary
instruments, (e.g., Imbens & Angrist, 1994; Kennedy, Lorch and Small, 2017.
Deterministic monotonicity states that for each subject, the level of the treatment
that a subject would take is a monotonic increasing function of the level of the IV,
i.e., requires a monotonic relationship between IV and treatment for each subject.
This assumption is unlikely to hold in most cases. A weaker version of mono-
tonicity, that of Stochastic Monotonicity, only requires a monotonic relationship
1 3
The Effect ofBody Image Perceptions onLife Satisfaction and…
between the IV and probability of treatment, hence it is easier to satisfy. Small
etal. (2017), show that under stochastic monotonicity, the IV method identifies a
weighted average of treatment effects.
Given the expectation that the instrument validity assumptions do not hold
precisely, in Sect. 4.5 (Further analysis), I implement the procedure suggested
by Nevo and Rosen (2012) which relaxes the key IV correlation assumption, by
replacing the zero-correlation assumption between the instrument and the unob-
served error term with an assumption related to the “sign” of the correlation,
thus allowing for the construction of IV bounds under weaker-than-traditional
assumptions.
2.5 Methodological Steps
The estimation method consists of three steps, outlined below.
(1) Derive a Plausible Range of Degree of Endogeneity of Perceived Body Image
Using the proposed instrument (whose relevance and validity is later tested), I
estimated Extended Regression Models2 (ERM) with endogenous covariates in
Stata, using maximum likelihood estimation. Such models, developed after Heck-
man’s (1976) original work on causal inference models (see also, Rubin, 1974;
Woolridge, 2010), incorporate both selection on observables and unobservables.
The model to be estimated, is as follows:
The outcome is the wellbeing indicator considered, determined by other covari-
ates (vector x1), the potentially endogenous covariate w1 (i.e., perceived body
image), and error term ε1. Having identified suitable instrument/s (x2), the poten-
tially endogenous covariate is determined by the instrument (x2), along with vari-
ables in vector x1, and error term ε2. Unobservables are represented by the error
terms in the four equations.
The error covariance matrix is:
(1)
y=(
x
1),β+w
1β1+
𝜀
1
(2)
w1=(
x
1),β2+
x
2
𝛾
1+
𝜀
2
2 Extended regression models (ERMs), implemented using Stata, are a specific class of models that
address complications that arise frequently in data including endogenous covariates, sample selec-
tion, and non-random treatment assignment. ERM models are similar to IV regression models and are
more flexible with respect to the complications that they address, but rely on previously identified valid
instrument/s.
C.Sakellariou
1 3
i.e., from this model, besides coefficient estimates, one can estimate point estimates
of correlation in error terms (unobservables) between the two equations (i.e., ρ12)
and associated confidence intervals. A negative error correlation estimate indicates
that unobservables which tend to increase body image satisfaction occur with unob-
servables which tend to decrease SWB; hence, effect estimates derived from an esti-
mator which do not correct for endogeneity biases will be downward biased. The
opposite is the case for positive error correlation estimates.
(2) Test Exclusion Restrictions at Plausible Degree of Endogeneity Using KLS
Excluded instruments need to be correlated with the endogenous regressor, but
uncorrelated with the outcome. The exclusion restrictions of potentially suitable instru-
ments were formally tested using the recently developed kinky least-squares (KLS)
approach by Jan Kiviet (Kiviet, 2020; Kripfganz & Kiviet, 2021); these restrictions
were previously considered untestable. KLS yields: (1) statistical inference on the
validity of exclusion restrictions regarding candidate external instruments, for a plausi-
ble range of endogeneity correlations, using the point estimate and confidence interval
of endogeneity correlations derived in step (1); and (2) instrument free effect estimates
at various values of the range of plausible degree of endogeneity. These estimates come
with substantially narrower confidence intervals compared to the IV estimates.
(3) Derive IV and KLS Estimates of the Effect of Perceived Body Image on Outcome
Using KLS regressions, an (instrument free) effect estimate associated with body
image perceptions can be derived, using as plausible range of degree of endogeneity,
the confidence interval around the derived point estimate of degree of endogeneity
of perceived body image. The KLS effect estimate can then be compared to the cor-
responding IV estimate. When the validity of instrument exclusion restrictions is
supported (step 2), the KLS estimate will generally mirror the IV estimate; however,
when the test comes with low p-values (less support of validity of exclusion restric-
tions), the KLS estimate is the preferred estimate, as it is adjusted for the specific
value of plausible degree of endogeneity of perceived body image.
2.6 Determinants andCorrelates ofSWB
In specifying the model for estimation, besides the variable of interest (perceived
body image), one needs to consider relevant observed determinates and correlates
of SWB, based on evaluative and emotional/affective SWB theories (i.e., how deter-
minants/correlates influence wellbeing),3 depending on the dimension of SWB used.
Σ=[
𝜎2ρ12
ρ12 1
]
3 Most of the empirical studies examine how determinants or correlates influence evaluative and/or emo-
tional dimensions of wellbeing.
1 3
The Effect ofBody Image Perceptions onLife Satisfaction and…
One also needs to consider the study population used (i.e., adults, elderly, adoles-
cents, or other age groups, etc.). A review of the literature (e.g., Das etal., 2020),
suggests the following broad categories of determinants/corelates of SWB irrespec-
tively of study population (see also, Diener etal., 1999): (1) Demographic character-
istics; (2) Geography; (3) Socioeconomic status/material resources; (4) Health and
disabilities; (5) Personality and temperament; (6) Social connections and support;
(7) Culture and religion.
Given that the study population in this study 15-year-old students, additional cor-
relates relate to school environment, such as sense of belonging in school and support
from parents (as part of social connections and support), adolescents’ interests, such
as interest/use of Information and Communication Technology (ICT), and free time
available for leisure (net of instruction and study time). Finally, since the variable of
interest is perceived body image, body mass index (BMI) should be controlled for.
2.7 Measures
2.7.1 Wellbeing Measures
Life Satisfaction When assessing LS and perceived quality of life, OECD opted to
focus on satisfaction (rather than the evaluative) aspect, using the 11-point (0–10)
LS scale based on the question: “How satisfied are you with your life overall these
days?” The rationale is that this question is easier to understand and less intrusive
for adolescent students than the life evaluation question (OECD, 2019).
SWB‑Positive Affect The survey contains a continuous scale, derived from three sur-
vey questions on frequency of positive emotions in a four-point Linkert scale (rang-
ing from “Never” to Always”): “How often do you feel Joyful”; “How often do you
feel Cheerful”; and “How often do you feel Happy.”
2.7.2 Potentially endogenous covariate ofinterest
Perceived Body Image The body image continuous index in PISA 2018 contains
negative and positive values and was derived from questions given on a four-point
Linkert scale (from “Strongly disagree” to Strongly agree”): “I like my look just the
way it is”; “I consider myself to be attractive”; “I like my body”; and “I like the way
my clothes fit me.” Reported scale reliabilities (Cronbach’s alpha) ranged by coun-
try, from 0.86 to 0.92.
2.7.3 Other Covariates
Several other student and school variables were considered as controls in a prelimi-
nary investigation:
Individual characteristics
C.Sakellariou
1 3
Demographic Characteristics and Geography Possible relevant fixed characteristics,
besides gender, include the 15-year old’s immigrant/minority status, and location.
Self‑Reported Health 15-year-olds were asked to assess their health as “Excellent”,
“Good”, “Fair”, or “Poor.
Body Mass Index (BMI) The index was derived from questions on self-reported
height in cm and weight in kgs. Using this index, I derived a categorical variable
with BMI ranges for being underweight, of normal weigh, overweight, and obese.
Socioeconomic Status and its Components The economic, social, and cultural sta-
tus index (ESCS) in PISA was derived from three indices: highest parental occupa-
tion, highest parental education, and student reports on home possessions, including
books in the home. From the home possessions index, five sub-indices were derived:
wealth possessions, cultural possessions, home educational resources, and informa-
tion and communication technology (ICT) resources.
Free Time in a Day Derived using information on school instruction time and stu-
dent’s out of school hours of study time.
Interest in/Use of ICT The ICT Familiarity Questionnaire in the survey contains
several variables on familiarity and usage of ICT in, or outside the school, derived
using IRT scaling.
Academic Achievement I used students’ average combined PISA score (reading,
math, and science) as a summary measure of academic achievement. Performing
well in tests, besides academic ability, may be associated with other observed or
unobserved attributes, such as how hard-working a student is, which may be cor-
related with wellbeing.
Perceived Resilience/Self‑Efficacy The resilience/self-efficacy index was derived
from answers to questions such as: “I usually manage one way or the other”, “I feel
proud that I have accomplished things”, “I feel that I can handle many things at a
time”, “My belief in myself gets me through hard times”, and “When I am in a diffi-
cult situation, I can usually find my way out of it.” Stronger agreement vs. disagree-
ment to each question contributes to more positive vs, more negative values in the
index.
Perceptions of Social Relationships The social dimension questions in the question-
naire relate to students’ perceptions on how they relate to their school’s environment
(sense of belonging in school), as well as their perception of relationships outside
the school. The following items were given in a 4-point Linkert scale: (a) Satisfac-
tion with friendship networks in general: How satisfied are you with the friends you
have? (b) Six items related to sense of connectedness/belonging in school: “I make
friends easily at school”; “other students seem to like me”; “I feel like I belong at
1 3
The Effect ofBody Image Perceptions onLife Satisfaction and…
school”; “I feel lonely at school”; “I feel awkward and out of place in my school”;
and “I feel like an outsider (or left out of things) at school”; and (c) Number of
friends. The set of six items related to sense of connectedness/belonging in school
exhibit high reliability for measuring the same construct (α = 0.806). This is also the
case when these six items are combined with the item on satisfaction with friend-
ship networks in general (α = 0.803). However, adding “number of friends” in the
group containing items under (a) and (b) severely reduces Cronbach’s alpha statistic.
Therefore, I used Principal Component Analysis (PCA) to derive a continuous index
on students’ perceptions of quality of social relationships, using the items in (a) and
(b). The index contains negative and positive values, with more positive values indi-
cating higher satisfaction with social relationships.
Perceptions of Emotional Support from Parents Derived from answers to questions
in a 4-point Linkert scale (Strongly Agree to Strongly Disagree): “My parents sup-
port my educational efforts”; “My parents support me when I am facing difficul-
ties”; and “My parents encourage me to be confident.”
2.7.4 School characteristics
Public/Private School Public vs Private (independent or government dependent)
school.
Single Sex School Single-sex school vs co-ed school.
Learning Environment Relevant derived indices include, students’ assessment of
school disciplinary climate, teacher support, practice of teacher-directed instruction,
and perceived teacher feedback.
3 Results
3.1 Estimation Sample andSummary Statistics
Observed gender differences in mean characteristics are generally statistically sig-
nificant (TableA1 in Appendix). Boys report higher LS and positive emotions than
girls. Boys are more satisfied with their body image compared to girls (as generally
reported in the literature), although based on BMI, a lower proportion of boys have
a BMI in the normal range. A much higher proportion of boys assess their health as
excellent compared to girls. Boys report more interest in ICT and substantially more
use of ICT in social interactions compared to girls. Finally, perceived resilience/self-
efficacy of boys is higher than that of girls, while girls report more emotional sup-
port from parents.
C.Sakellariou
1 3
3.2 Model Specification
The vector of retained controls (besides body image perceptions), includes location
(town, city, large city, compared to village), immigrant status self-assessment health
(Good, Fail, Poor, compared to Excellent), BMI range (underweight, overweight,
obese, compared to normal range), socioeconomic status, hours of free time in a
day, interest in ICT/use of ICT in social interaction, school ownership, single-sex
(vs co-ed) school, disciplinary climate (as a proxy for learning environment), per-
ceptions of resilience, perceptions of quality of social relationships, perceptions of
emotional support from parents, and average achievement score in PISA tests. The
model also controls for country-school fixed effects. The outcome as well as all con-
tinuous indices among the controls were standardized (mean = 0 and SD = 1).
While expected signs of associations between some of the covariates and out-
come (such as for body image perceptions, health, resilience, social relationships,
and emotional support from parents) are intuitive, there is no theoretically based
expected sign for certain covariates, such as academic achievement, interest/use of
ICT, and having more free/discretionary time.
Empirical studies examining the relationship between adolescents’ objective
indicators of academic achievement and SWB have shown conflicting results and
the reasons are not fully understood (Crede etal., 2015), with different conclusions
drawn from associations on a national level compared to associations on an indi-
vidual level (Suldo etal., 2006). From related empirical evidence, the relationship
between ICT and SWB can be influenced by whether one refers to access to such
technologies, vs intensity of use of computers, the internet, and mobile phones, and
for what purpose (e.g., Castellacci & Schwabe, 2020; Maiti & Awasthi, 2020; Zhong
etal., 2022). The relationship between having more free time and SWB can also be
nuanced, depending on amount of free time and how it is spent (i.e., hobbies, work-
ing out, etc., vs watching TV, using the internet, etc.), and findings can vary (e.g.,
Sharif, 2021).
3.3 The effect ofbody image onLife Satisfaction andSubjective Wellbeing
3.3.1 Naïve Model Estimates
Column 1 in Tables1 and 2 contains the Ordinary Least Squares (OLS) estimates
of the association between wellbeing measure and covariates, assuming selection
on observables only (i.e., that we can account for factors resulting in selection by
controlling only for observed characteristics). Model estimation used the PISA sur-
vey design and weights (except when deriving KLS estimates). With both outcome
and all continuous covariates standardized at mean = 0 and SD = 1, effect estimates
(β) associated with body image and other continuous covariates are interpreted
accordingly.
1 3
The Effect ofBody Image Perceptions onLife Satisfaction and…
Table 1 Effect estimates with Life Satisfaction as Outcome
Outcome: Life Satisfaction
(stand.)
MALES FEMALES
OLS IV KLS OLS IV KLS
Body Image (stand.) 0.126
(0.016) 0.169
(0.055) 0.177
(0.007) 0.124
(0.020) 0.169
(0.076) 0.200
(0.006)
Town (vs village) –0.012
(0.036)
–0.013
(0.036)
–0.012
(0.015)
–0.033
(0.038)
–0.034
(0.037)
–0.023
(0.014)
City (vs village) 0.010
(0.037)
0.011
(0.038)
–0.029
(0.015)
–0.003
(0.048)
–0.003
(0.048)
–0.060
(0.014)
Large city (vs village) –0.025
(0.049)
–0.025
(0.050)
–0.069
(0.017)
–0.013
(0.048)
–0.015
(0.048)
–0.072
(0.016)
Immigrant –0.191
(0.034)
–0.196
(0.033)
–0.152
(0.015)
–0.219
(0.045)
–0.224
(0.044)
–0.172
(0.014)
Health condition: Good
(vs excellent)
–0.203
(0.029)
–0.192
(0.032)
–0.186
(0.012)
–0.145
(0.039)
–0.136
(0.044)
–0.195
(0.011)
Health condition: Fair (vs
excellent)
–0.487
(0.062)
–0.470
(0.066)
–0.434
(0.021)
–0.363
(0.054)
–0.347
(0.062)
–0.461
(0.017)
Health condition: Poor (vs
excellent)
–0.655
(0.192)
–0.637
(0.189)
–0.727
(0.051)
–0.458
(0.201)
–0.431
(0.202)
–0.770
(0.044)
Underweight (vs normal
weight)
–0.078
(0.074)
–0.075
(0.074)
–0.042
(0.025)
–0.198
(0.076)
–0.207
(0.077)
–0.106
(0.025)
Overweight (vs normal
weight)
0.006
(0.045)
0.019
(0.044) 0.048
(0.017)
–0.012
(0.052)
0.002
(0.057) 0.051
(0.018)
Obese (vs normal weight) 0.131
(0.061) 0.154
(0.068)
0.034
(0.025)
0.098
(0.064)
0.118
(0.069) 0.081
(0.027)
Socioeconomic Status
index
–0.015
(0.016)
–0.017
(0.016)
–0.019
(0.006)
0.021
(0.021)
–0.020
(0.022)
–0.023
(0.006)
5−9 hours of free time in a
day (vs < 5)
–0.099
(0.034)
–0.099
(0.035)
–0.054
(0.013)
–0.012
(0.036)
–0.010
(0.037)
–0.060
(0.012)
≥ 10 hours of free time in
a day (vs < 5)
–0.247
(0.033)
–0.245
(0.043)
–0.141
(0.018)
–0.198
(0.049)
–0.193
(0.048)
–0.097
(0.019)
Average PISA score –0.094
(0.019)
–0.091
(0.020)
–0.045
(0.007)
–0.043
(0.031)
–0.040
(0.031)
0.011
(0.007)
Interest in ICT index –0.029
(0.014)
–0.029
(0.014)
–0.039
(0.006)
–0.068
(0.020)
–0.066
(0.021)
–0.067
(0.006)
Public school –0.077
(0.037)
–0.074
(0.036)
–0.012
(0.013)
–0.029
(0.045)
–0.029
(0.045)
–0.023
(0.012)
Single-sex school –0.276
(0.062)
–0.280
(0.061)
–0.119
(0.023)
–0.022
(0.067)
–0.024
(0.069)
–0.076
(0.020)
School disciplinary cli-
mate index 0.065
(0.017) 0.065
(0.017) 0.063
(0.005) 0.085
(0.019) 0.084
(0.019) 0.055
(0.005)
Resilience/self-efficacy
index 0.081
(0.018) 0.075
(0.020) 0.106
(0.006) 0.161
(0.017) 0.150
(0.024) 0.125
(0.005)
Social relationships index 0.098
(0.020) 0.090
(0.021) 0.134
(0.006) 0.140
(0.025) 0.132
(0.028) 0.154
(0.006)
Parental emotional support
index 0.100
(0.020) 0.096
(0.020) 0.115
(0.006) 0.196
(0.020) 0.191
(0.021) 0.171
(0.006)
Constant 0.725
(0.074) 0.712
(0.077) 0.456
(0.029) 0.662
(0.078) 0.656
(0.080) 0.493
(0.026)
F-statistic [p-value] 52.7 [0.000] 43.3 [0.000] - 85.9 [0.000] 79.9 [0.000] -
C.Sakellariou
1 3
Table 1 (continued)
Outcome: Life Satisfaction
(stand.)
MALES FEMALES
OLS IV KLS OLS IV KLS
First stage:
Body Image: school level
mean
-0.776
(0.055)
- - 0.758
(0.051)
-
Town (vs village) –0.055
(0.041)
0.013
(0.015)
City (vs village) –0.056
(–0.042)
0.015
(0.015)
Large city (vs village) –0.035
(0.044)
0.003
(0.016)
Immigrant 0.135
(0.030) 0.116
(0.023)
Health condition: Good
(vs excellent)
–0.247
(0.029)
–0.200
(0.012)
Health condition: Fair (vs
excellent)
–0.379
(0.047)
–0.320
(0.017)
Health condition: Poor (vs
excellent)
–0.325
(0.167)
–0.537
(0.049)
Underweight (vs normal
weight)
–0.068
(0.065) 0.190
(0.033)
Overweight (vs normal
weight)
–0.260
(0.040)
–0.327
(0.016)
Obese (vs normal weight) –0.525
(0.059)
–0.429
(0.026)
Socioeconomic Status
index 0.044
(0.013)
–0.041
(0.005)
5−9 hours of free time in a
day (vs < 5)
0.017
(0.032)
–0.029
(0.011)
≥ 10 hours of free time in
a day (vs < 5)
–0.028
(0.050)
–0.031
(0.023)
Average PISA score –0.073
(0.018)
–0.053
(0.008)
Interest in ICT index 0.009
(0.016)
–0.027
(0.006)
Public school 0.063
(0.032)
–0.015
(0.014)
Single-sex school 0.041
(0.036)
0.035
(0.033)
School disciplinary cli-
mate index
–0.004
(0.016) 0.013
(0.006)
Resilience/self-efficacy
index 0.133
(0.016) 0.233
(0.005)
Social relationships index 0.173
(0.017) 0.175
(0.006)
Parental emotional support
index 0.078
(0.018) 0.103
(0.006)
Constant 0.273
(0.068) 0.166
(0.032)
1 3
The Effect ofBody Image Perceptions onLife Satisfaction and…
Table 1 (continued)
KLS estimate based on postulated endogeneity of body image of –0.052 for males and –0.053 for females
The model controls for fixed effects in all models. Bold indicates significance at the 5% level or lower
Outcome: Life Satisfaction
(stand.)
MALES FEMALES
OLS IV KLS OLS IV KLS
F-statistic [p-value] 289.4 [0.000] 278.1 [0.000]
Partial R-sq. 0.071 0.062
Endogeneity test for Body Image:
Wu-Hausman F-value
[p-value]
0.946 [0.331] 0.902 [0.342]
N25,117 25,117 25,117 26,616 26,616 26,616
The naive model estimates of the association between body image perceptions
and LS are generally modest and effect sizes do not differ by gender (β = 0.126; 95%
CI: [0.096, 0.156] for males and β = 124; 95% CI: [0.084, 0.165] for females). Based
on confidence intervals, the corresponding estimates for the association between
body image and Positive Affect are higher for girls (β = 0.168; 95% CI: [0.128,
0.208], compared to β = 0.090; 95% CI: [0.055, 0.125] for boys).
3.3.2 Other Covariates
Self-assessed physical health is a major determinant of LS; reporting fair or poor
health (compared to excellent health) is associated with about 0.5–0.65 SD lower
LF for boys and 0.35–0.45 SD lower LF for girls. The corresponding associations
between health condition and SWB-Positive Affect are 0.3–0.4 SD lower Positive
Affect for boys and 0.4–0.6 SD for girls. Male and female students of immigrant
background report lower LS (β = -0.2), but no statistically significant association
between immigrant status and SWB-Positive Affect was found. The association
between socioeconomic status and both outcomes is negative but weak, when sta-
tistically significant. Reporting more discretionary time (≥ 10 h of free time in a
day compared to < 5), is negatively associated with both LS and SWB-Positive
Affect, suggesting that unobserved attributes are at play. Interest in ICT displays
a small negative association with LS, while use of ICT for social interaction dis-
plays a small positive association with SWB-Positive Affect for boys. Among school
level variables, attending a single-sex school is negatively associated with LS for
boys (β = -0.28), but no such association with SWB-Positive Affect was found. The
association between perceptions of resilience and LS is stronger for girls (β = 0.16,
vs β = 0.08 for boys), while its association with SWB-Positive Affect is of similar
magnitude for girls and boys (β = 0.12–0.14). The association between perceptions
of social relationships and SWB-Positive Affect (β = 0.20–0.22) is stronger than the
corresponding association with LS (β = 0.1–0.14). Perceived emotional support from
parents is more strongly associated with LS of girls (β = 0.2, vs β = 0.1 for boys).
Finally, scoring higher in PISA tests is negatively associated with LS and SWB-
Positive Affect, but effect sizes are small (β <—0.1).
C.Sakellariou
1 3
Table 2 Effect estimates with SWB-Positive Affect as outcome
Outcome: Positive Affect
(stand.)
MALES FEMALES
OLS IV KLS OLS IV KLS
Body Image (stand.) 0.090 0.071 0.069 0.168 0.203 0.190
(0.018) (0.063) (0.007) (0.020) (0.099) (0.007)
Town (vs village) -0.010 -0.010 -0.030 0.065 0.063 0.065
(0.044) (0.043) (0.017) (0.045) (0.044) (0.016)
City (vs village) -0.031 -0.042 -0.039 0.038 0.037 0.012
(0.048) (0.048) (0.017) (0.052) (0.051) (0.016)
Large city (vs village) -0.084 -0.084 -0.077 0.055 0.054 0.039
(0.052) (0.052) (0.021) (0.053) (0.053) (0.019)
Immigrant -0.059 -0.057 -0.028 -0.003 -0.005 0.028
(0.041) (0.041) (0.021) (0.045) (0.046) (0.020)
Health condition: Good
(vs excellent) -0.238 -0.243 -0.212 -0.201 -0.194 -0.168
(0.036) (0.039) (0.013) (0.036) (0.040) (0.013)
Health condition: Fair (vs
excellent) -0.320 -0.328 -0.343 -0.393 -0.381 -0.361
(0.061) (0.066) (0.024) (0.056) (0.065) (0.020)
Health condition: Poor (vs
excellent)
-0.420 -0.428 -0.645 -0.628 -0.610 -0.588
(0.234) (0.228) (0.061) (0.145) (0.148) (0.048)
Underweight (vs normal
weight)
-0.084 -0.085 -0.024 -0.155 -0.164 -0.050
(0.102) (0.102) (0.030) (0.083) (0.084) (0.030)
Overweight (vs normal
weight)
0.038 0.132 0.029 0.082 0.094 0.108
(0.037) (0.057) (0.020) (0.052) (0.063) (0.022)
Obese (vs normal weight) 0.138 0.128 0.116 0.162 0.179 0.077
(0.079) (0.088) (0.035) (0.083) (0.096) (0.039)
Socioeconomic Status
index -0.048 -0.047 -0.032 -0.028 -0.025 -0.031
(0.018) (0.018) (0.006) (0.016) (0.018) (0.006)
5–9h of free time in a day
(vs < 5) -0.084 -0.084 -0.054 -0.072 -0.071 -0.088
(0.037) (0.037) (0.015) (0.035) (0.036) (0.013)
≥ 10h of free time in a
day (vs < 5) -0.201 -0.202 -0.151 -0.186 -0.183 -0.126
(0.048) (0.049) (0.020) (0.078) (0.075) (0.020)
Average PISA score -0.095 -0.096 -0.065 -0.037 -0.035 -0.017
(0.021) (0.022) (0.008) (0.024) (0.025) (0.008)
Use of ICT as social inter-
action index 0.033 0.033 0.019 -0.024 -0.023 0.005
(0.016) (0.016) (0.007) (0.020) (0.020) (0.007)
Public school 0.051 0.052 0.017 -0.072 -0.072 0.055
(0.037) (0.037) (0.014) (0.043) (0.043) (0.013)
1 3
The Effect ofBody Image Perceptions onLife Satisfaction and…
Table 2 (continued)
Outcome: Positive Affect
(stand.)
MALES FEMALES
OLS IV KLS OLS IV KLS
Single sex school 0.043 0.046 0.060 -0.070 0.068 0.065
(0.085) (0.085) (0.067) (0.087) (0.088) (0.040)
School disciplinary cli-
mate index
0.018 0.018 0.023 0.028 0.028 0.007
(0.019) (0.019) (0.007) (0.017) (0.017) (0.006)
Resilience/self-efficacy
index 0.138 0.140 0.143 0.124 0.116 0.110
(0.016) (0.018) (0.007) (0.018) (0.030) (0.006)
Social relationships index 0.222 0.226 0.236 0.195 0.188 0.212
(0.019) (0.023) (0.007) (0.023) (0.026) (0.007)
Parental emotional support
index 0.066 0.068 0.099 0.110 0.106 0.109
(0.019) (0.019) (0.007) (0.019) (0.021) (0.007)
Constant 0.406 0.412 0.213 0.383 0.376 0.322
(0.077) (0.081) (0.031) (0.084) (0.089) (0.028)
F-statistic [p-value] 48.7 [0.000] 47.5 [0.000] - 61.7 [0.000] 57.5 [0.000] -
First stage:
Body Image: school level
mean 0.785 0.757
(0.058) (0.056)
Town (vs village) 0.003 0.007
(0.044) (0.040)
City (vs village) -0.055 0.023
(0.045) (0.040)
Large city (vs village) -0.026 -0.010
(0.047) (0.042)
Immigrant 0.155 0.093
(0.040) (0.043)
Health condition: Good
(vs excellent) -0.247 -0.193
(0.031) (0.031)
Health condition: Fair (vs
excellent) -0.383 -0.309
(0.049) (0.048)
Health condition: Poor (vs
excellent)
-0.303 -0.434
(0.148) (0.146)
Underweight (vs normal
weight)
-0.072 0.233
(0.047) (0.072)
Overweight (vs normal
weight) -0.251 -0.318
(0.043) (0.044)
C.Sakellariou
1 3
Table 2 (continued)
Outcome: Positive Affect
(stand.)
MALES FEMALES
OLS IV KLS OLS IV KLS
Obese (vs normal weight) -0.506 -0.439
(0.064) (0.082)
Socioeconomic Status
index 0.043 -0.050
(0.013) (0.014)
5–9h of free time in a day
(vs < 5)
0.017 -0.026
(0.033) (0.030)
≥ 10h of free time in a
day (vs < 5)
-0.029 -0.032
(0.052) (0.044)
Average PISA score -0.070 -0.045
(0.019) (0.018)
Interest in ICT (social
interaction) index
0.016 0.024
(0.017) (0.018)
Public school 0.060 -0.008
(0.034) (0.035)
Single-sex school 0.089 0.033
(0.085) (0.077)
School disciplinary cli-
mate index
-0.004 0.008
(0.019) (0.016)
Resilience/self-efficacy
index 0.136 0.236
(0.017) (0.015)
Social relationships index 0.170 0.181
(0.018) (0.018)
Parental emotional support
index 0.078 0.097
(0.020) (0.017)
Constant 0.257 0.200
(0.070) (0.068)
F-value [p-value] - 267.9 [0.000] - - 236.1 [0.000] -
Partial R-sq 0.072 0.053
Endogeneity test for Body Image:
Wu-Hausman F-value
[p-value]
0.50 [0.480] 2.22 [0.137]
N18,637 18,637 18,637 19,380 19,380 19,380
KLS estimate based on postulated endogeneity of body image of 0.022 for males and -0.041 for females
The model controls for country fixed effects in all models. Bold indicates significance at the 5% level or
lower
1 3
The Effect ofBody Image Perceptions onLife Satisfaction and…
3.4 Considering Endogeneity ofBody Image Perceptions
Ignoring potential endogeneity of covariates of interest introduces biases in esti-
mated effects. The direction of the bias depends on the sign and strength of the cor-
relation between the endogenous covariate and unobserved attributes. When the true
effect of the covariate on the outcome is expected to be positive (as is the case with
all three covariates treated as potentially endogenous), and the covariate is positively
correlated with unobserved attributes, the coefficient from the naive regression will
be upward biased; on the other hand, if it is negatively correlated the true effect will
be larger (the coefficient from the naive regression will be downward biased).
As a first step, I estimated Extended Regression Models (ERMs) by gender, with
body image as an endogenous covariate and body image at the school level as instru-
ment. Given a suitable instrument, ERMs produce estimates mirroring IV estimates;
they also provide a point estimate and confidence interval of the degree of endoge-
neity of the potentially endogenous covariate (error correlations between first stage
and second-stage equations). In the model with LS as outcome, point estimates and
confidence intervals of degree of endogeneity of body image were -0.052 (95% CI:
-0.16, 0.06) for males; and -0.053 (95% CI: [-0.2, 0.1]) for females; with SWB-
Positive Affect as outcome the corresponding estimates were 0.02 (95% CI: [-0.11,
0.15]) for males and -0.041 (95% CI: [-0.22, 0.14]) respectively for females.
Using the survey design and replication weights, I derived IV estimates of the
effect of body image on each of the outcomes (along with coefficient estimates for
other covariates), which are given in column 2 of Tables1 and 2. First stage statis-
tics show that the instrument is strong in with both outcomes. Tests of endogene-
ity of body image with LS as outcome suggest a modest degree of endogeneity4 in
both male and female regressions, with the null hypothesis that body image is exog-
enous rejected at the 5% level of significance (p-value = 0.33 for males and 0.34 for
females]. With SWB-Positive Affect as outcome, the null is accepted at high p-value
for males, while for females there is weak evidence of endogeneity of body image
[p-value = 0.13]. Effect size (β) reflects the standardized change in outcome for one
SD increase in body image satisfaction. With LS as outcome, effect sizes from IV
regressions are identical for males and females (β = 0.169; 95% CI: [0.06, 0.27] for
males; and β = 0.169; 95% CI: [0.02, 0.32] for females). These estimates are modest
in size and somewhat larger than those from OLS regressions. With SWB-Positive
Affect as outcome, the effect for females is three times that for males (β = 0.203;
95% CI: [0.02, 0.40], vs. β = 0.071; 95% CI: [-0.055, 0.197]). However, this differ-
ence is statistically significant only at p-value of about 0.15. Thus, more positive
body image likely affects the frequency of positive emotions more for girls than
boys. These IV estimates rely on the validity of the instrument used, which has not
yet been tested.
4 The endogeneity test is valid insofar as the instrument is valid.
C.Sakellariou
1 3
3.4.1 KLS Estimation andInstrument testing
The KLS method allows (a) testing of exclusion restrictions of instruments (sin-
gle or a set) given a plausible range of endogeneity correlations and (b) deriving
(unweighted) instrument-free coefficient estimates at specific values of degree of
endogeneity of the potentially endogenous covariate, along with (unweighted) IV
estimates. Graphs from deriving IV and KLS effect estimates using the KLS method
and testing the exclusion restrictions for each outcome by gender are given in the
Appendix. The instrument used in the IV estimation was tested around the earlier
derived point estimates of degree of endogeneity. The tests are given graphically
(lower part of Graphs 1 and 2), using the associated p-values to test the null hypoth-
esis of validity of exclusion restrictions. The top part of the graphs shows the KLS
estimates of the effect of body image on the outcome at various points in the grid of
plausible values of degree endogeneity of body image, along with the IV estimate.
The third column in Tables1 and 2 contains the KLS coefficient estimates of the
effect of body image on the outcome at the point estimates of the degree of endo-
geneity derived earlier, along with the coefficient estimates of the other covariates.
With LS as outcome, the null hypothesis is accepted for males at very high
p-value; hence, the (unweighted) KLS point estimate5 (β = 0.177; 95% CI: [0.165,
0.190]) is very close to the (unweighted) IV estimate (β = 0.184; 95% CI: [0.135,
0.233]). The corresponding IV estimate using survey weights, reported in Table1
(β = 0.169) is also of similar magnitude. For females, the hypothesis that the instru-
ment is valid is accepted only at a p-value of about 0.10. The KLS estimate, adjusted
at the point estimate of degree of endogeneity of body image (β = 0.200; 95% CI:
[0.188, 0.212]), is somewhat smaller that the unweighted IV estimate (β = 0.244;
95% CI: [0.197, 0.291]. Comparing KLS estimates by gender, the much narrower
95% confidence intervals associated with KLS estimates allows for the conclusion
that the effect size estimate of 0.2 for females is larger than the corresponding male
estimate of 0.177; however, given that standard errors associated with KLS esti-
mates come from unweighted regressions, this conclusion may be unwarranted.
With SWB-Positive Affect as outcome, the validity of the instrument is accepted
at high p-values for both males and females. Unweighted KLS point estimates for
males and females (β = 0.069; 95% CI: [0.054, 0.083] for males; and β = 0.190;
95% CI: [0.176, 0.204] for females) are essentially identical to the corresponding
unweighted IV estimates (β = 0.073; 95% CI: [0.02, 0.13] for males; and β = 0.189;
95% CI: [0.131, 0.248] for females). Comparing effect sizes between males and
females, the estimate for females, at about 0.2 SD higher SWB-Positive Affect for
one SD increase in body image satisfaction, is three times the small effect size esti-
mate of about 0.07 SD of males.
5 Since KLS estimates are unweighted, they are inconsistent. Corresponding unweighted IV estimates
are given for comparison.
1 3
The Effect ofBody Image Perceptions onLife Satisfaction and…
3.5 Further analysis
3.5.1 Assuming Imperfect Instrument
I derived IV bounds under assuming that the instrument does not hold precisely,
i.e., relaxing the zero covariance assumption (cov(z,u) = 0), following the Nevo
and Rosen (2012) approach.6 Given an endogenous variable of interest, x, and an
instrument, z, instead of assuming that ρzε = 0, it is assumed that the instrument has
(weakly) the same direction of correlation with the omitted error term as endoge-
nous covariate (i.e., ρxερzε ≥ 0). If the instrument is less endogenous than the endog-
enous variable of interest (i.e., ρxε ≥ ρzε), then this amounts to essentially consider-
ing an “imperfect” instrumental variable.
Bounds from IV estimation assuming imperfect instruments (using Stata’s
imperfectiv module) are reported in TablesA2a and A2b in the Appendix. Given
that the correlation between the endogenous variable and the imperfect instru-
ment is positive, only one-sided intervals are reported. The estimates were derived
when the assumption that the instrument is less correlated with the error term
than the original endogenous variable (Assumption 4 in the paper) does not hold.
The tables contain estimates after considering both assumptions about the corre-
lation between the endogenous variable and the unobservable error (positive vs.
negative).
The estimates reported in tablesA2a and A2b are consistent with the main IV
regression results for both outcomes. With LS as outcome, the effect estimates,
and confidence interval bounds support the earlier finding of no gender differ-
ence in the effect of perceived body image on LS. With SWB-Positive affect as
outcome, the estimates support the earlier finding that effect estimates larger for
female participants with the gender difference statistically significant. Considering
the earlier reported finding pointing to a negative correlation between error terms
in the outcome and first-stage regressions, provides further support for the main
conclusions.
3.5.2 Omitting Covariates
The main results were derived after conditioning for all available observable
covariates, some of which are potentially endogenous. I investigated the sensi-
tivity of findings and main conclusions to omitting (a) potentially endogenous
covariates and (b) all covariates other than body image perceptions. Specifically,
I looked at: (1) validity of exclusion restrictions and (2) sensitivity of estimates of
the effect of body image perceptions on the two outcomes. I found that first, with-
out additional controls, or controlling only for fixed student and school charac-
teristics, the coefficient estimate associated with body image perceptions is sub-
stantially higher than when controlling for the full set of observable covariates.
6 Sarrias and Blanco (2020) used this partial identification strategy, in a context similar to this study.
C.Sakellariou
1 3
This is expected since the coefficient estimate partially reflects the effect of omit-
ted characteristics on outcome. Second, when other covariates are omitted, find-
ings with respect to degree of endogeneity and validity of exclusion restrictions
are very much like the main findings, i.e., exclusion restrictions are satisfied at a
plausible range of degree of endogeneity of body image. However, this is not the
case when partially controlling for observables, for example by controlling for
fixed characteristics only. These findings seem to support suggestions in the liter-
ature that the main reason for including covariates is that the assumptions related
to identification may be more plausible after conditioning on covariates, and that
instruments are sometimes valid only after conditioning for certain covariates
(Angrist etal., 2000; de Caisemartin, 2017). In addition, covariates make infer-
ence more precise.
4 Discussion
Earlier empirical research on SWB and its determinants/correlates varies in differ-
ent respects, such as theoretical and methodological framework, dimension of SWB
studied, type of data, population coverage, and model specification, hence, making
comparability of findings difficult. From the review of the literature by Das etal.
(2020), most studies have looked at one dimension of wellbeing (with the evaluative
dimension being the dominant one), with only one-third of studies looking at both
the evaluative (mainly the Satisfaction with Life Scale) and emotional (with Positive
and Negative Affect Scale the most common measure) dimensions. The great major-
ity of studies was cross-sectional (and more so in psychology), with none using an
experimental or quasi-experimental design. With respect to study population, nearly
half of the 105 studies reviewed were on adults or the elderly, nine on college stu-
dents and only three on adolescents. Only one study used a multi-country sample.
Furthermore, the range of determinants/correlates varied considerably; the majority
of studies considered basic demographics, socioeconomic status, and health-related
determinants, while a much smaller proportion of studies considered personality,
social connections/support, geography, religion and culture determinants and corre-
lates. Similar differences with respect to study population, methodology, and deter-
minants/correlates are evident in empirical studies on the link between perceived
body image and SWB.
Motivated by these weaknesses in earlier empirical research this study, with
adolescents as the study population, employed a quasi-experimental methodology,
intended to tease out gender specific effect sizes of body image perceptions on two
wellbeing indicators: one evaluative, using the 0–10 (overall) Satisfaction with Life
Scale, and the other on the emotional dimension of SWB (Positive Affect index).
The dataset contains information on almost all the theoretically relevant determi-
nants/correlates of adolescents’ SWB (in addition to perceived body image), hence,
allowing for a rich set of covariates in the specification of the model.
1 3
The Effect ofBody Image Perceptions onLife Satisfaction and…
Addressing the two research questions, the findings established that (1) there is
evidence of biases due to endogeneity of perceived body image, which are modest
in size; (2) substantial gender differences in effect estimates were found when the
outcome was the emotional dimension of SWB. With LS as outcome, bias-corrected
estimates are about 40–50% higher for both males and females. With SWB-Positive
Affect as outcome, bias-corrected estimates sharpen gender differences in effect
size (already present from the OLS regressions), with the revised female effect size
estimate about three times larger than the corresponding male estimate. Therefore,
the findings suggest that body image perceptions influence the emotional dimension
measure of SWB (frequency of positive emotions) substantially more for girls than
for boys. On the other hand, the effect on the evaluative dimension of wellbeing
(LS) is of similar magnitude for both girls and boys. It can be argued that the find-
ings on gender differences in effects by dimension of wellbeing studied make intui-
tive sense.
Bias-corrected effect sizes, while modest, are not negligible. The size of
the effect of one SD higher body image satisfaction on (standardized) LS, esti-
mated at about 0.2 SD increase in LS, corresponds to approximately 0.5 of a
step higher in the 0–10 Satisfaction with Life Scale. Similarly, for girls the size
of the effect of one SD higher body image satisfaction (estimated at about 0.2
SD higher Positive Affect), corresponds to about 10 points higher in the percen-
tile distribution of SWB-Positive Affect, when evaluated around the mean of the
distribution.
Comparison of findings to those in the empirical literature is difficult. This is
due to the small number of studies on adolescents, and a variety of differences in
scope, methodological approach, body image and outcome measures, model speci-
fication, sample size, and other differences. Earlier studies generally do not control
for more than a few observed characteristics, while samples are rarely large. Several
studies used body dissatisfaction as the measure of body image, while frequently the
outcome is depression/depressive symptoms or related outcomes, with a few using
SWB measures. Few studies were designed for estimation of effect sizes. Generally,
the objective of earlier studies was to establish an association between body image
and various outcomes, especially when longitudinal data were available. This study
differs in its objective and methodology, focusing on effect estimates which can be
assigned a quasi-causal interpretation, along with gender differences in the size of
these estimates. Given the above differences between this and earlier studies (and
scarcity of reported effect size estimates on adolescents), comparison of estimates
would not be meaningful. Studies on adults are somewhat more prevalent, and some
report magnitude of effects. Generally, effect sizes found are modest (as found in this
study) and similar in size by gender; however, this study documented larger effects
for female participants when the outcome is frequency of positive emotions. For
example, Davies etal. (2019) estimated hierarchical multiple regression models, con-
trolling for age, BMI, and psychological attributes, to derive associations between
two body image measures (body dissatisfaction and body appreciation) and LS. They
C.Sakellariou
1 3
found that body appreciation7 is a strong predictor of LS, and moderate effect sizes
for male and female participants; on the other hand, effect sizes for body dissatisfac-
tion were substantially smaller.
The approach outlined in this study can be used to investigate the sensitivity of effect
estimates in a variety of contexts involving psychological attributes and perceptions. For
example, in this study, the vector of determinants/correlates of wellbeing contains potentially
endogenous covariates other than perceived body image, i.e., perceptions of resilience, and
sense of belonging/social connections. I investigated biases due to potential endogeneity of
these variables and associated gender differences in effect estimates8 using the methodo-
logical steps outlined earlier. The findings suggest moderate biases associated with the naïve
model, and mostly in the female regressions. With LS as outcome, bias-corrected estimates
for effect of resilience are larger in the female regression (β = 0.25 vs. β = 0.16), but no sub-
stantial change for males. No significant differences between the OLS and IV/KLS estimates
of the effect of sense of belonging/social connections on LS were found for either males or
females. With SWB-Positive Affect as outcome, the bias-corrected effect of sense of belong-
ing/social connections is somewhat higher in the female regression (β = 0.3 vs. β = 0.2).
This study has limitations. It would be preferable to apply the quasi-experimental
approach used in this study with longitudinal data. Furthermore, the cross-country
data sample used for analysis covers the nine countries (from Europe, Latin Amer-
ica, East Asia, and Middle East), which participated in the wellbeing module of
PISA 2018; hence, the findings may not be generalizable in the sense that, had the
estimation sample been larger, findings may have been somewhat different.
5 Conclusion
I outlined and implemented a quasi-experimental approach to derive bias-corrected
estimates of the effect of perceived body image on the evaluative and emotional wellbe-
ing of adolescents, and associated gender differences in effect size. I used a large, multi-
country estimation sample, and a model specification accounting for most theoretically
relevant determinants/correlates of wellbeing. Biases associated with endogeneity of
perceived body image exist and are generally modest in size. Bias-corrected effect sizes
with the evaluative dimension of wellbeing as outcome (LS) are somewhat larger com-
pared to the naïve estimates and not substantially different by gender. Substantial gen-
der differences in effect size estimates on the emotional dimension of wellbeing (Posi-
tive Affect) were established, with the effect size for girls about three times larger than
for boys. The proposed methodology can be used to investigate causal effect in a variety
of contexts involving potentially endogenous variables of interest (such as psychologi-
cal attributes and perceptions) and various outcomes besides SWB (e.g., mental health).
7 However, another problem in comparing findings is that body appreciation and degree of satisfaction
with one’s body image (perceived body image), are distinct from one another; the body appreciation
scale in the referenced study is based on different survey questions than those used in deriving the body
image perception composite in this study.
8 While a more detailed investigation is outside the scope of the study, estimation results are available
upon request.
1 3
The Effect ofBody Image Perceptions onLife Satisfaction and…
Table 3 Weighted summary statistics by gender
Characteristic Male Female M-F Diff
Life Satisfaction (Ladder mean) 7.97 7.63 *
SWB-Positive Affect (index mean) 0.307 0.275 *
Body image (index mean) 0.215 0.032 *
Locality: Village (%) 27.7 28.1 -
Locality: Town (%) 23.7 22.3 *
Locality: City (%) 27.2 27.8 -
Locality: Large city (%) 21.4 21.8 -
Immigrant (%) 5.8 7.0 *
Health: Excellent (%) 47.7 31.3 *
Health: Good (%) 41.9 51.5 *
Health: Fair (%) 9.3 15.8 *
Health: Poor (%) 1.1 1.3 *
BMI range: Normal weight (%) 76.9 80.3 *
BMI range: Underweight (%) 5.5 3.1 *
BMI range: Overweight (%) 12.8 12.1 *
BMI range: Obese (%) 4.8 4.5 *
ESCS (index mean) -0.784 -0.883 *
< 5h of free time (%) 34.6 45.7 *
5–9h of free time (%) 57.3 49.1 *
> 9h of free time (%) 8.1 5.1 *
Interest in ICT (index mean) -0.015 -0.035 *
Use of ICT for social interaction (index mean) 0.272 0.022 *
Public school (%) 83.2 81.5 *
Single sex school (%) 1.5 2.7 *
School disciplinary climate (index mean) -0.120 -0.017 *
Resilience/self-efficacy (index mean) 0.104 -0.015 *
Social relationships composite (index mean) -0.015 -0.087 *
Emotional support from parents (index mean) -0.058 0.093 *
Average PISA test score 434.6 434.3 -
* indicates male–female difference is significant at the 5% level or lower
Appendix Table 3 Table 4 Table 5
Table 4 Nevo and Rosen (2012)’s Imperfect IV bounds—Males
Outcome Positive correlation Negative correlation
Body image: Upper
bound estimate
Body image:
Upper bound CI
Body image: Lower
bound estimate
Body image:
Lower bound
CI
Life satisfaction 0.124 0.153 0.166 0.096
Positive Affect 0.073 0.129 0.091 0.053
C.Sakellariou
1 3
Tests of Exclusion Restrictions of Instruments Graph.1 Graph.2
Table 5 Nevo and Rosen (2012)’s Imperfect IV bounds – Females
Outcome Positive correlation Negative correlation
Body image: Upper
bound estimate
Body image: Upper
bound CI
Body image
Lower bound
estimate
Body image:
Lower bound
CI
Life satisfaction 0.121 0.156 0.165 0.087
Positive Affect 0.166 0.202 0.209 0.131
Outcome: Life Satisfaction
MALE
SF
EMALES
Graph 1 Outcome: Life Satisfaction
1 3
The Effect ofBody Image Perceptions onLife Satisfaction and…
Outcome: SWB-Positive Affect
MALE
SF
EMALES
Graph 2 Outcome: SWB-Positive Affect
C.Sakellariou
1 3
Acknowledgements Not applicable.
Authors Contributions This is a single author manuscript.
Data Availability The datasets analysed during the current study are available fromhttps:// www. oecd. org/
pisa/ data/ 2018d ataba se/
Declarations
Ethics Approval and Consent to Participate Not applicable. The data was accessed under a standard End
User Licence arrangement for an academic research project as the data are fully anonymised.
Consent for Publication No consent for publication was required as the data are freely available for aca-
demic research from oecd.org.
Competing Interests The author declares that they have no financial or non-financial competing interests.
References
Angrist, J., Graddy, K., & Imbens, G. (2000). The interpretation of instrumental variables estimators
in simultaneous equations models with application to the demand for fish. The Review of Eco-
nomic Studies, 67(3), 499–527.
Bearman, S. K., & Stice, E. (2008). Testing a gender additive model: The role of body image in ado-
lescent depression. Journal Abnormal Child Psychology, 36(8), 1251–1263. https:// doi. org/ 10.
1007/ s10802- 008- 9248-2
Bornioli, A., Lewis-Smith H., Slater A., & Bray, I. (2020). Body dissatisfaction predicts the onset of
depression among adolescent females and males: a prospective study. Epidemiology & Community
Health, 75(4) https:// doi. org/ 10. 1136/ jech- 2019- 213033
Castellacci, F., & Schwabe, H. (2020). Internet, unmet aspirations and the U-shape of life. PLoS One,
15(6), e0233099. https:// doi. org/ 10. 1371/ journ al. pone. 02330 99
Crede, J., Wirthwein, L., McElvany, N., & Steinmayr, R. (2015). Adolescents’ academic achievement and
life satisfaction: The role of parents’ education. Frontiers in Psychology, 6, 52. https:// doi. org/ 10.
3389/ fpsyg. 2015. 00052
Das, K., Jones-Harrell, C., Yingling, F., Ramaswami, A., Orlove, B., & Botchwey, N. (2020). Under-
standing subjective well-being: Perspectives from psychology and public health. Public Health
Reviews, 41, 25. https:// doi. org/ 10. 1186/ s40985- 020- 00142-5
Davies, L.L., Fawler, S.A., Best, L.A., and Both, L.E. (2020). The role of body image in the prediction of
life satisfaction and flourishing in men and women. Journal of Happiness Studies, 505–524. https://
doi. org/ 10. 1007/ s10902- 019- 00093-y
de Caisemartin, C. (2017). Tolerating defiance? Local average treatment effects without monotonicity.
Quantitative Economics, 8(2017), 367–396. https:// doi. org/ 10. 3982/ QE601
Diener, E. (1984). Subjective well-being. Psychological Bulletin, 95(3), 542–575. https:// doi. org/ 10.
1037/ 0033- 2909. 95.3. 542
Diener, E., Suh, E., Lucas, R., & Smith, H. (1999). Subjective Weil-Being: Three Decades of Progress.
Psychological Bulletin, 125(2), 276–302. https:// doi. org/ 10. 1037/ 0033- 2909. 125.2. 276
Gilman, R., & Huebner, E. S. (2003). A Review of Life Satisfaction Research with Children and Adoles-
cents. School Psychology Quarterly, 18(2), 192–205. https:// doi. org/ 10. 1521/ scpq. 18.2. 192. 21858
Gilman, R., & Huebner, E. S. (2006). Characteristics of Adolescents Who Report Very High Life
Satisfaction. Journal of Youth and Adolescence, 35(3), 311–319. https:// doi. org/ 10. 1007/
s10964- 006- 9036-7
Gilman, R., Huebner, S., & Laughlin, J. E. (2000). A First Study of the Multidimensional Students’ Life
Satisfaction Scale with Adolescents. Social Indicators Research, 52, 135–160. https:// doi. org/ 10.
1023/A: 10070 59227 507
1 3
The Effect ofBody Image Perceptions onLife Satisfaction and…
Heckman, J. (1976). The Common Structure of Statistical Models of Truncation, Sample Selection and
Limited Dependent Variables and a Simple Estimator for Such Models. In: Sanford V. Berg (Ed.)
Annals of Economic and Social Measurement, 5(4), 475–492 http:// www. nber. org/ chapt ers/ c10491
Holmqvist Gattario, K., Frisen, A., & Anderson-Fye, E. (2014). Body Image and Child Well-Being. In:
A. Ben Arieh, F. Casas, I. Frønes, J. E. Korbin (Eds)., Handbook of child well-being, 2409-2436,
Springer
Huebner, E. S. (1994). Preliminary development and validation of a multidimensional life satisfaction
scale for children. Psychological Assessment, 6(2), 149–158. https:// doi. org/ 10. 1037/ 1040- 3590.6.
2. 149
Huebner, E. S., Laughlin, J. E., Ash, C., & Gilman, R. (1998). Further Validation of the Multidimen-
sional Students’ Life Satisfaction Scale. Journal of Psychoeducational Assessment, 16(2), 118–134.
https:// doi. org/ 10. 1177/ 07342 82998 01600 202
Huebner, E. S., Suldo, S. M., & Gilman, R. (2006). Life Satisfaction. In G. G. Bear & K. M. Minke
(Eds.), Children’s needs III: Development, prevention, and intervention (pp. 357–368). National
Association of School Psychologists.
Imbens, G., & Angrist, J. (1994). Identification and Estimation of Local Average Treatment Effects.
Econometrica, 62(2), 467–475. https:// doi. org/ 10. 2307/ 29516 20
Kennedy, E., Lorch, S., & Small, D. (2018). Robust Causal Inference with Continuous Instruments Using
the Local Instrumental Variable Curve. Journal of the Royal Statistical Society Series b: Statistical
Methodology, 81(1), 121–143. https:// doi. org/ 10. 1111/ rssb. 12300
Kiviet, J. F. (2020). Testing the impossible: Identifying exclusion restrictions. Journal of Econometrics,
218, 294–316. https:// doi. org/ 10. 1016/j. jecon om. 2020. 04. 018
Kripfganz, S., & Kiviet, J. F. (2021). kinkyreg: Instrument-free inference for linear regression models
with endogenous regressors. The Stata Journal, 21(3), 772–813. https:// doi. org/ 10. 1177/ 15368
67X21 1045
Lee, B.J., and Yoo, M.S. (2015). Family, School and Community Correlates of Children’s Subjective
Well-being: An International Comparative Study. Child Indicators Research, 8(1) https:// doi. org/ 10.
1007/ s12187- 014- 9285
Lemma, P., Borraccino, A., Berchialla, P., and Dalmasso, P. Well-being in 15-year-old adolescents: a
matter of relationship with school. Journal of Public Health, 37(4), 573–580 https:// doi. org/ 10.
1093/ pubmed/ fdu095
Lousdal, M. L. (2018). An introduction to instrumental variable assumptions, validation and estimation.
Emerging Themes in Epidemiology, 15(1), 1–7. https:// doi. org/ 10. 1186/ s12982- 018- 0069-7
Maiti, D., & Awasthi, A. (2020). ICT Exposure and the Level of Wellbeing and Progress: A Cross Country
Analysis. Social Indicators Research, 147, 311–343. https:// doi. org/ 10. 1007/ s11205- 019- 02153-5
Mond, J., van den Berg, P., Boutelle, K., & Hannan, P. (2011). Obesity, Body Dissatisfaction, and Emo-
tional Well-Being in Early and Late Adolescence: Findings from the Project EAT Study. Journal of
Adolescent Health, 48(4), 373–378. https:// doi. org/ 10. 1016/j. jadoh ealth. 2010. 07. 022
Nevo, A., and Rosen, A. M. (2012). Identification with imperfect instruments. The Review of Economics
and Statistics, 94(3), 659–671. https:// www. jstor. org/ stable/ 23261 470
OECD. (2019). PISA 2018 Technical Report: Chapter16. France.
Rassen, J., Brookhart, M., Glynn, R., Mittleman, M., & Schneeweiss, S. (2009). Instrumental variables I:
Instrumental variables exploit natural variation in nonexperimental data to estimate causal relation-
ships. Journal of Clinical Epidemiology, 62, 1226–1232. https:// doi. org/ 10. 1016/j. jclin epi. 2008. 12.
005
Rubin, D. B. (1974). Estimating causal effects of treatments in randomized and nonrandomized studies.
Journal of Educational Psychology, 66(5), 688–701. https:// doi. org/ 10. 1037/ h0037 350
Rutkowski, L. (2016). Introduction to special issue on quasi-causal methods. Large Scale Assessments in
Education, 4(8), 1–6. https:// doi. org/ 10. 1186/ s40536- 016- 0023-5
Sarias, M., & Blanco, A. (2022). Bodyweight and human capital development: Assessing the impact
of obesity on socioemotional skills during childhood in Chile. Economics & Human Biology, 47,
101190. https:// doi. org/ 10. 1016/j. ehb. 2022. 101190
Sharif, M. (2021). Having Too Little or Too Much Time is Linked to Lower Subjective Well-Being. Jour-
nal of Personality and Social Psychology, 121(4), 933–947. https:// doi. org/ 10. 1037/ pspp0 000391
Small, D., Tan, Z., Ramsahai, R., Lorch, S., & Brookhart, A. M. (2017). Instrumental Variable Estima-
tion with a Stochastic Monotonicity Assumption. Statistical Science, 32(4), 561–579. https:// doi.
org/ 10. 1214/ 17- STS623
C.Sakellariou
1 3
Smolak, L., & Cash, T. (2011). Future challenges for body image science, practice, and prevention. In
T. F. Cash & L. Smolak (Eds.), Body image: A handbook of science, practice, and prevention (pp.
471–478). The Guilford Press.
Suldo, S., & Huebner, E. S. (2004). Does life satisfaction moderate the effects of stressful life events
on psychopathological behavior during adolescence? School Psychology Quarterly, 19(2), 93–105.
https:// doi. org/ 10. 1521/ scpq. 19.2. 93. 33313
Suldo, S., Riley, K., & Shaffer, E. J. (2006). Academic Correlates of Children and Adolescents’ Life
Satisfaction. School Psychology International, 27(5), 567–582. https:// doi. org/ 10. 1177/ 01430 34306
073411
Suldo, S. M., Minch, D., & Hearon, B. V. (2015). Adolescent life satisfaction and personality charac-
teristics: Investigating relationships using a five-factor model. Journal of Happiness Studies:
An Interdisciplinary Forum on Subjective Well-Being, 16(4), 965–983. https:// doi. org/ 10. 1007/
s10902- 014- 9544-1
Swami, V., Weis, L., Barron, D., & Furnham, A. (2017). Positive body image is positively associated with
hedonic (emotional) and eudaimonic (psychological and social) well-being in British adults. Jour-
nal of Social Psychology, 158(5), 541–552. https:// doi. org/ 10. 1080/ 00224 545. 2017. 13922 78
Veenhoven, R. (1988). The utility of happiness. Social Indicators Research, 20, 333–354. https:// doi. org/
10. 1007/ BF003 02332
Woolridge, J. (2010). Econometric Analysis of Cross Section and Panel Data (2nd ed.). MIT Press.
Yates, A., Edman, J., & Aruguete, M. (2004). Ethnic difference in BMI and body/self-dissatisfaction
among Whites, Asian subgroups, Pacific Islanders, and African Americans. Journal of Adolescent
Health, 34, 300–307. https:// doi. org/ 10. 1016/j. jadoh ealth. 2003. 07. 014
Zhong, J., Wu, W., & Zhao, F. (2022). The impact of Internet use on the subjective well-being of Chinese
residents: From a multi-dimensional perspective. Frontiers in Psychology. https:// doi. org/ 10. 3389/
fpsyg. 2022. 950287
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under
a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of such publishing agreement and
applicable law.
A preview of this full-text is provided by Springer Nature.
Content available from Child Indicators Research
This content is subject to copyright. Terms and conditions apply.