Beruflich Dokumente
Kultur Dokumente
BRIEF REPORT
Up, Not Down: The Age Curve in Happiness From Early Adulthood to
Midlife in Two Longitudinal Studies
Nancy L. Galambos, Shichen Fang,
Harvey J. Krahn, and Matthew D. Johnson
Margie E. Lachman
Brandeis University
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
University of Alberta
Happiness is an important indicator of well-being, and little is known about how it changes in the early
adult years. We examined trajectories of happiness from early adulthood to midlife in 2 Canadian
longitudinal samples: high school seniors followed from ages 18 43 and university seniors followed
from ages 2337. Happiness increased into the 30s in both samples, with a slight downturn by age 43 in
the high school sample. The rise in happiness after high school and university remained after controlling
for important baseline covariates (gender, parents education, grades, self-esteem), time-varying covariates known to be associated with happiness (marital status, unemployment, self-rated physical health),
and number of waves of participation. The upward trend in happiness runs counter to some previous
cross-sectional research claiming a high point in happiness in the late teens, decreasing into midlife. As
cross-sectional designs do not assess within-person change, longitudinal studies are necessary for
drawing accurate conclusions about patterns of change in happiness across the life span.
Keywords: happiness, midlife, growth curve modeling, longitudinal, early adulthood
Measuring Happiness
The social science literature on the meaning and measurement
of happiness is voluminous, riddled with differing opinions, and
extends back at least to the 1930s. Our approach to conceptualizing
1664
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
1665
18 and a high point around age 50. The Midlife in the U.S.
longitudinal study (adults ages 2575 followed for 10 years) found
positive affect was stable from the mid-20s into the late 30s,
declined from the 30s into the 40s, rose from the 50s into the 60s,
and was highest at ages 60 69 (Lachman et al., 2015). Using the
German Socio Economic Panel Study, Baird, Lucas, and Donnellan (2010) reported that ratings of happiness were largely stable
throughout adulthood with a downturn in later life. The longitudinal U.S. National Health and Nutrition Examination Survey
found little change in positive affect across 9 years in participants
ages 2574 at baseline (Costa et al., 1987). Positive affect was also
stable between young adulthood and midlife in a 23-year longitudinal study (Charles, Reynolds, & Gatz, 2001). These conflicting
results, along with the near absence of longitudinal studies tracking
teens into midlife, suggest conclusions about the shape of the
age happiness curve in the first half of adult life are premature.
1666
Table 1
Means and Standard Deviations of Happiness and Time-Varying Covariates in Two Samples
Wave 1
M
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
18
2.39
SD
Wave 3
M
0.58
2.48
Wave 4
SD
19
20
0.55
2.50
SD
Wave 5
M
22
0.55
2.57
Wave 6
SD
25
0.52
2.55
SD
Wave 7
M
32
0.55
2.66
SD
43
0.50
2.63
0.53
0.01
0.09
0.00
0.00
3.91
1.21
961968
0.05
0.21
1.31
2.71
4.09
1.18
653657
0.07
0.26
1.13
2.30
3.95
1.01
542543
0.21
0.40
0.57
1.48
4.01
.92
497500
0.42
0.49
0.86
2.22
3.89
1.00
399400
0.67
0.47
0.13
0.93
3.95
0.96
504505
0.76
0.42
0.09
0.92
3.70
1.01
387403
23
24
25
27
30
37
2.55
0.55
0.14
0.35
0.00
0.00
4.08
1.19
567574
Wave 2
SD
2.60
0.53
0.20
0.40
1.00
2.03
4.29
1.08
451453
2.58
0.53
0.26
0.44
0.70
1.64
3.94
0.95
417418
2.63
0.49
0.43
0.50
0.43
1.38
4.01
0.90
388389
2.68
0.48
2.72
0.47
0.61
0.49
0.48
1.48
4.06
0.91
352354
0.78
0.16
4.05
0.42
1.16
0.84
Method
High School Sample
Participants (N 968; at baseline, 47% women, 80% born in
Canada, 15% non-White, and 26% with at least one universityeducated parent) were from the Edmonton Transitions Study
(ETS), a 25-year longitudinal study that began in spring 1985 with
983 Grade 12 students (age 18) who completed questionnaires in
classrooms in six schools representing working- and middle-class
neighborhoods in a large western city in Canada. The baseline
sample was representative of western Canadian urban youth born
in 1967 on race, immigration status, and parents education.
Follow-up questionnaires were mailed in 1986 (Wave 2; age 19;
n 665), 1987 (Wave 3; age 20; n 547), 1989 (Wave 4; age 22;
n 503), and 1992 (Wave 5; age 25; n 404) only to previous
wave respondents. In 1999 (Wave 6), a telephone survey targeted
all baseline participants (age 32; n 509; response of 52% after
14 years). In 2010 (Wave 7), telephone and web surveys targeted
all baseline participants (age 43; n 405; 41% response after 25
years). Half (51%) had participated in all waves (for details, see
previous publications; Chow, Galambos, & Krahn, in press; Johnson, Galambos, & Krahn, 2014; Vargas Lascano, Galambos,
Krahn, & Lachman, 2015).
344
Possible range: 15. not
Measures
Means and standard deviations for happiness across time and for
time-varying covariates are presented in Table 1. Happiness was
assessed with an item similar to single-item measures labeled as
happiness in the GSS (e.g., Blanchflower & Oswald, 2008; Oishi,
Kesebir, & Diener, 2011), the 2000 Social Capital Benchmark
Survey (Subramanian et al., 2005), and the Seattle Longitudinal
Study (Hoppmann et al., 2011): Thinking about your life in
general, how happy are you with your life? Responses were coded
as 1 (not very happy at all), 2 (somewhat happy), or 3 (very happy).
Single-item happiness measures are sensitive to objective circumstances in individuals lives (Dolan, Peasgood, & White, 2008).
We examined construct validity by correlating our happiness item
with the felt depressed item from the Center for Epidemiologic
StudiesDepression Scale (CES-D; Radloff, 1977); concurrent
correlations for six waves in which both items were available
ranged from .38 to .50. In 2010, the happy CES-D item was
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
University Sample
In parallel with the first six waves of data collection for the high
school sample, 589 graduating baccalaureate degree students age
30 and younger (48% women) from the five largest faculties at a
large western Canadian university completed mail questionnaires
in 1985 (n 589; M age 23.3, SD 1.97, range 20.130.3
years); every third graduating student from an alphabetized list in
these faculties was contacted. The baseline sample is representative of the Canadian population of university graduates from these
five disciplines in the mid-1980s. Follow-ups were completed in
1986 (n 458), 1987 (n 421), 1989 (n 392), and 1992 (n
357). In 1999, a telephone survey targeted all baseline participants
(n 349, 59% of original sample); 281 had participated in all
previous waves (for details, see Galambos & Krahn, 2008). After
excluding individuals with no score for happiness at any wave or
no self-esteem or grades score at baseline, the final sample size
was 574.
Gender, parents education (M .36, SD .48), and selfesteem (M 4.11, SD .56) were measured as in the high school
sample. Self-reported senior year grades (M 6.99, SD .77)
were on the universitys 9-point scale where 9 A or A and
13 F. Time was coded in the same manner as the first six waves
of the high school sample.
Attrition analyses found that participants in 1999 were not
different from the baseline sample on gender. Nor did the 1999
1667
Plan of Analysis
For both samples, three multilevel models (Raudenbush & Bryk,
2002; Singer & Willett, 2003) were constructed using HLM 7.01.
First, an unconditional means model with no covariates at Level 1
(within-person level) and no predictors at Level 2 (betweenpersons level) partitioned the proportion of total variance in happiness to within-person and between-persons factors. Second, an
unconditional growth model estimated the intercept (happiness at
baseline) and the linear time slope or rate of change in happiness
across time; a quadratic time slope was also tested. Third, a final
model (a) examined between-persons predictors (gender, parents
education, senior year grades, self-esteem) of the intercept and
linear time slope, (b) controlled for number of waves of participation on the intercept, and (c) added time-varying covariates
(marital status, unemployment, physical health) to Level 1. The
intercept in all models was estimated as randomly varying across
persons, as was the linear time slope. The quadratic time slope and
all time-varying covariates were specified as nonrandomly varying. Full-information maximum likelihood was used to generate
parameter estimates and to preserve cases containing withinperson missing values. Dichotomous predictors at baseline (gender, parents education) were uncentered, and grades, self-esteem,
and waves of participation were grand centered. Time-varying
covariates (marriage, unemployment, physical health) were person
centered.
Results
The unconditional means model determined that for the high
school sample, 34% of the total variance in happiness was between
persons and 66% was within person (38% and 62%, respectively,
in the university sample). In the high school sample, the unconditional growth model revealed a linear increase in happiness (B
.027, p .05) and a quadratic slope (B .001, p .05). On
average, happiness rose in the first 14 years of the study, with a
faster rate of change between ages 18 and 25 than between ages 25
and 32, followed by a slight decrease by age 43. The inflection
point was at 35.71 years. The correlation between the intercept and
linear time slope was .61: happiness rose faster for the less happy
at baseline.
The unconditional growth model for the university sample also
revealed a linear increase in happiness (B .021, p .05) but no
quadratic change (B .001, p .05); the intercept and linear
time slope were correlated at .68. The model-implied quadratic
trajectories for the high school sample (panel a) and the linear
trajectories for the university sample (panel b) appear in Figure 1,
illustrating the variability both across and within persons. There
was sufficient variation in the intercept and linear slope to justify
exploration of between-persons predictors of both parameters in
the next set of models for both samples.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
1668
Discussion
We modeled growth trajectories in happiness from early adulthood to midlife among high school graduates followed from ages
18 43 and university graduates followed from ages 2337. In the
former sample, happiness increased from age 18 into the early 30s
and then turned slightly downward by the early 40s, a quadratic
trend evident in the unadjusted analyses (unconditional growth
model) and remaining after controlling for baseline (gender, parents education, grades, self-esteem) and time-varying (marriage,
unemployment, self-rated physical health) predictors as well as
number of waves of participation. In the university sample, happiness increased to age 37, with or without controls. Thus, both
data sets document a linear increase in happiness from the late
teens and early 20s into the early (high school sample) and late
(university sample) 30s, contrasting with arguments (e.g., Blanchflower & Oswald, 2008; Stone et al., 2010) that happiness decreases from early adulthood into midlife. Both studies also contradict Ulloa et al.s (2013) conclusion that the concavity
hypothesis, which necessarily includes an upward trend from
early adulthood to midlife, can most likely be dismissed (p. 240).
The slight downturn between ages 32 and 43 in our high school
sample did not reverse the earlier 14-year gains in happiness.
We believe the core difference between our results and others
lies in our longitudinal design. Cross-sectional studies (e.g.,
Blanchflower & Oswald, 2008; Stone et al., 2010) naturally confound age and cohort differences and cannot logically lead to
conclusions about intraindividual change, although some authors
are comfortable in drawing them. In both our longitudinal samples,
two thirds of total variability in happiness was within-person and
one third was between persons, demonstrating considerable intraindividual change. In contrast, all of the variability in crosssectional data is necessarily between persons; conclusions about
within-person change based solely on between-persons data are
fundamentally flawed. Additionally, as shown by Frijters and
Beattons (2012) research on life satisfaction, selection effects
inherent in cross-sectional data significantly affect results and are
much reduced in longitudinal analyses that account for the fact that
happier people are more likely to experience happiness-increasing
events such as marriage. Even accelerated designs, in which longitudinal slices are strung together to generate a life span picture of
1669
Table 2
Final Models Predicting Happiness Trajectories in Two Samples
High school sample (n 968)
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
B
Fixed effects
Intercept
Participation
Gender
Parent education
Grades
Self-esteem
Linear term
Gender
Parent education
Grades
Self-esteem
Quadratic term
Time-varying covariates
Married
Unemployed
Physical health
Random effects (variance)
Intercept
Linear term
2.470
0.029
0.103
0.093
0.030
0.315
0.010493
0.004
0.005
0.000
0.011
0.000342
0.184
0.010
0.089
ES
4.989
3.858
3.103
1.900
15.264
0.16
0.12
0.10
0.06
0.44
1.916
2.076
0.107
6.212
0.06
0.07
0.00
0.20
7.073
1.895
9.065
0.16
0.04
0.20
0.087
0.000
ES
0.881
2.719
1.664
0.531
11.299
0.04
0.11
0.07
0.02
0.43
0.412
1.251
0.968
3.296
0.02
0.05
0.04
0.14
2.072
3.870
5.323
0.06
0.10
0.14
0.093
0.000
p .05.
change (e.g., Frijters & Beatton, 2012) are not ideal, as they
assume individuals tracked longitudinally early in life will follow
the same path as individuals whose growth trajectories speak to
later portions of the life span, again confounding age and cohort
differences.
Our evidence of substantial intraindividual change across early
adulthood and into midlife conforms with theoretical predictions
of plasticity in human development (Baltes, 1987) and with research demonstrating mental health improvements from the teens
to early 20s (e.g., Meadows et al., 2006). It is important to
remember that, although the average trajectory of happiness increased into the 30s, individual trajectories (and random effects for
the intercept and linear slope) reveal differences across persons in
intraindividual change (see Figure 1), similar to the diversity
observed by Hoppmann et al. (2011). This is another reason why
general patterns of growth could differ from one study to the next.
For example, stable trajectories in positive affect found for earlier
portions of the adult life span (Charles et al., 2001; Costa et al.,
1987) likely obscure numerous trajectories flowing in opposite
directions. We know from our results that diversity in individual
growth trajectories begins early as a function of gender, socioeconomic background, and self-esteem, predictors that influence the
level of happiness and alter the rate of growth. Additional diversity
emerges from important life events such as changes in marital
status, unemployment, and physical health, which raise or lower
happiness when they occur. There may be an average happiness
trajectory across the life span that longitudinal studies can discern but there is no single trajectory.
With respect to predictors of happiness, women were happier in
both samples, a small but significant effect. Subramanian et al.
(2005) also found higher happiness in women in a sample of
24,118 U.S. residents. Diener et al. (1999) concluded that women
and men generally do not differ in global happiness but where
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
1670
References
Baird, B. M., Lucas, R. E., & Donnellan, M. B. (2010). Life satisfaction
across the lifespan: Findings from two nationally representative panel
studies. Social Indicators Research, 99, 183203. http://dx.doi.org/10
.1007/s11205-010-9584-9
Baltes, P. B. (1987). Theoretical propositions of life-span developmental
psychology: On the dynamics between growth and decline. Developmental Psychology, 23, 611 626. http://dx.doi.org/10.1037/0012-1649
.23.5.611
Blanchflower, D. G., & Oswald, A. J. (2008). Is well-being U-shaped over
the life cycle? Social Science & Medicine, 66, 17331749. http://dx.doi
.org/10.1016/j.socscimed.2008.01.030
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
1671