Money does not always buy happiness, but are richer

people less happy in their daily lives? It depends on how

you analyze income

Laura Kudrnaa, Kostadin Kushlevb

a
Institute of Applied Health, University of Birmingham, Murray Learning Centre, B15 2TT,
[email protected], +44 121 414 4388
b
Department of Psychology, Georgetown University, 306N White-Gravenor Hall, 37 th and O
Streets, N.W., Washington DC, 20057, [email protected], 202-687-4042

Conflicts of Interest and Source of Funding: This work was supported by a London School of
Economics PhD Studentship awarded to Dr Laura Kudrna. The funder played no role in the
study design, data collection and analysis, decision to publish, or preparation of the
manuscript.

Address correspondence to: Laura Kudrna, Institute of Applied Health, University of

Birmingham, Murray Learning Centre, B15 2TT, [email protected], +44 121 414 4388
Abstract
Do people who have more money feel happier during their daily activities? Some prior
research has found no relationship between income and daily happiness when treating income
as a continuous variable in OLS regressions, although results differ between studies. We re-
analyzed existing data, treating household income as a categorical variable and using lowess
and spline regressions to explore non-linearities. Our analyses reveal that these
methodological decisions provide new insights into the relationship between income and
happiness. We find some evidence that higher income is associated with less happiness and
no substantive benefit to higher household income in the US after $35-40K and in Germany
after €14-18K. Not all analytic approaches generate the same conclusions, which may explain
discrepant results.
Does more money make people happier? al, 2008; Aknin et al, 2012). Relatedly,
The answer to this question has psychological research suggests that people
implications for how individuals live their with lower socio-economic status have a
lives and societies are structured. It is more interdependent sense of self (Snibbe
often assumed that more income brings & Markus, 2005; Stephens et al., 2007). It
more happiness. In many aspects of policy, is, therefore, possible that people high in
upward income mobility is encouraged, income have lower wellbeing because they
and poverty can result in exclusion, experience less of the internal ‘warm glow’
stigmatization and discrimination by (Andreoni et al., 1990) benefit that comes
institutions and members of the public. along with valuing social relationships and
More income provides people with group membership. In theory, therefore,
capabilities and opportunities to consume there are reasons to suppose that high
more and thus satisfy more of their income has both benefits and costs for
preferences, meet their desires and obtain wellbeing, and empirical evidence can
more of what they want and need. These inform the debate about when and whether
are all reasons to assume that higher these different perspectives are supported.
income will bring greater happiness – or,
at least, that low income will bring low Empirical evidence on income and
happiness. happiness

Some research challenges the assumption The standard finding in existing literature
that earning more should lead to greater is that higher income predicts greater
happiness. First, because people expect that happiness, but with a declining marginal
more money should make them happier, utility (Dolan et al., 2008): that is, higher
people may feel less happy when their high income is most closely associated with
expectations are not met (Graham & happiness among those with the least
Pettinato, 2002; Nickerson et al., 2003). income, and is least closely associated
Second, since the 1980s in many developed with happiness for those with the most
countries, the well educated have had less income. Recently, this finding has been
leisure time than those who are not (Aguiar qualified by studies showing that the
& Hurst, 2007) and people living in high- relationship between income and
earning and well-educated households happiness depends on how happiness is
report feeling more time stress and conceptualized and measured: as an
dissatisfaction with their leisure time overall evaluation of one’s life or as daily
(Hamermesh & Lee, 2007; Nikolaev, emotional states (Kahneman & Deaton,
2018). Evidence also shows that people 2010; Killingsworth, 2021). In this vein,
with higher incomes spend more time alone authors Kushlev, Dunn, and Lucas (2015)
(Bianchi & Vohs, 2016). The lower found no relationship between income and
quaility and quantity of leisure and social daily happiness in the American Time Use
time of people with higher incomes may, in Survey (ATUS). This finding was
turn, negatively impact their happiness. replicated in the German Socioeconomic
Panel Survey (GSEOP) by Hudson, Lucas,
At the same time, some – but not all – Donnellan, and Kushlev (2015), and in
evidence suggests that working class another analysis of the ATUS by Stone,
individuals tend to be more generous and Schneider, Krueger, Schwartz, and Deaton
empathetic than more affluent individuals (2018).
(Piff et al., 2010; Kraus et al., 2010;
Balakrishnan et al, 2017), and such Some research has focused specifically on
kindness towards others has been the effect of high income on happiness.
associated with higher wellbeing (Dunn et Kahneman and Deaton (2010) conducted
regression analyses using a Gallup sample categorical variable. We compare these
of United States residents, finding that results to GSOEP data where we re-code
annual income beyond ~$75K was not the original continuous measure of income
associated with any higher daily emotional into categorical quantiles. To further
wellbeing. Income beyond ~$75K, explore non-linearities in the relationship
however, predicted better life evaluations. between income and happiness, we also
Using a self-selecting sample of conduct local linear ‘lowess’ and spline
experiential data in the United States, regression analyses.
Killingsworth (2021) conducted piecewise
regressions and found no evidence of We chose to re-analyze these data because
satiation or turning points. Jebb, Tay, the ATUS and GSOEP provide nationally
Diener, and Oishi (2018) fit regression representative data on people’s feelings as
spline models to global Gallup data, experienced during specific ‘episodes’ of
showing that the ‘satiation point’ in daily the day after asking them to reconstruct
experiences found by Kahneman and what they did during the entire day. Thus,
Deaton was also apparent in other compared to data from Gallup, which
countries. Unlike Kahneman and Deaton, measures affect ‘yesterday’, measurements
however, Jebb and colleagues also found in the ATUS are more grounded in specific
evidence of satiation in people’s life experiences, and therefore, less subject to
evaluations, and even some evidence for recall bias (Kahneman et al, 2004). And
‘turning points’ – whereby richer people unlike Gallup, which uses more crude,
evaluated their lives as worse than some of dichotomous (‘yes-no’) response scales,
those with lower incomes. This pattern of ATUS measures happiness along a
findings could partly depend on the choice standard 7-point Likert-type scale. In the
of analytic strategy. In analyses of the GSOEP, we were also able to analyze data
same dataset with lowess regression, from the Experience Sampling
researchers found no evidence of satiation Methodology (ESM), which asks people
or turning points in the relationship how they are feeling during specific
between income and people’s life episodes during the day and, as such, is
evaluations (Sacks, Stevenson, & Wolfers, even more grounded in specific
2012; Stevenson & Wolfers, 2012). These experiences.
conflicting results suggest that the effect of
analytic strategy on results deserves a The original ATUS income variable—
closer examination. family income—contains 16 uneven
categories (see Table 1). For example,
The current research Category 11 has a range of ~$10K,
whereas Category 14 has a range of
We propose that the relationship between ~$25K. The increasingly larger categories
income and happiness may depend not are designed to reflect declining marginal
only on how happiness is measured, but utility as an innate quality of income.
also on how income is measured and Based on this, Kushlev and colleagues
analyzed. In this paper, we focus on non- (2015) analyzed income as a continuous
linearities and re-analyze the ATUS data variable using the original uneven
used by Kushlev, Dunn, and Lucas (2015) categories. Continuous scales, however,
and Stone and colleagues (2018), as well assume equal intervals between scale
as the GSOEP data used by Hudson and points—a strong assumption to make for
colleagues (2016). Specifically, while the relatively arbitrary rate of change in
Kushlev and colleagues analyzed income the category ranges. Is increasing one’s
as a continuous variable in the ATUS, we income from $20,000 to $25,000 really
treat income the way it was measured: as a equidistant to increasing it from $35,000 to
$40,000 (Table 1)? And can we really income that is continuous in its original
assume, for example, that adding $5,000 of form. Whether analyzing this income
additional income to $35,000 is the same measure in its raw original form or in
as adding $10,000 of additional income to transformed log and quadratic forms, a
$40,000? Recognizing this issue, income null relationship with happiness was
researchers have adopted alternative observed. This approach, however, does
strategies. For example, Stone and not consider whether there might be non-
colleagues (2018) took the midpoints of linear/log/quadratic turning or satiation
each category of income, and then log- points at higher levels of income – an issue
transformed it. Thus, they transformed the also applicable to previous analyses of
categorical measure of income into a ATUS (Kushlev et al, 2015; Stone et al.,
continuous measure. This approach 2018). This is important because there are
produced results for happiness consistent theoretically both benefits and costs to
with the findings of Kushlev and achieving higher levels of income that
colleagues. could occur at various levels of income,
however, this possibility has not yet been
Both the increasing ranges of the income fully explored in ATUS or GSOEP data.
scale itself and its log-transformations
reflect an assumed declining marginal In sum, past research using ATUS has
utility of income: They treat a given treated categorically measured income as a
amount of income increase at the higher continuous variable, either assuming
end of the income distribution as having equidistance between scale points or
less utility than the same amount at the attempting to create equidistance through
lower end of the distribution. But by statistical transformations. By doing so,
subsuming income’s declining utility in however, researchers may have
its very measurement (or transformation statistically accounted for the very utility
thereof), it becomes difficult to interpret a of income for happiness that they are
null relationship with happiness. In other trying to test. In both ATUS and GSOEP,
words, we might not be seeing a declining the question of whether there might be
marginal utility of income reflected on satiation and/or turning points at higher
happiness because the income variable levels of income has not been fully
itself reflects its declining utility. considered. The present research explores
whether treating income as a categorical
Even when the income variable itself does variable in both ATUS and GSOEP would
not reflect its declining utility, a null replicate past findings or reveal novel
relationship between income and daily insights, focusing on possible non-
experiences of happiness has been linearities in the relationship between
observed. Hudson and colleagues (2015) income and happiness.
used GSOEP, which contains a measure of
Table 1. The original categories of income in the ATUS
family income measure with number of individuals in
each income category in the ATUS 2010, 2012 and 2013
wellbeing modules.
Group N
Income range
number (individuals)
1 Less than $5,000 883
2 $5,000 to $7,499 645
3 $7,500 to $9,999 903
4 $10,000 to $12,499 1,221
5 $12,500 to $14,999 1,096
6 $15,000 to $19,999 1,773
7 $20,000 to $24,999 2,005
8 $25,000 to $29,999 1,989
9 $30,000 to $34,999 2,044
10 $35,000 to $39,999 1,809
11 $40,000 to $49,999 2,959
12 $50,000 to $59,999 2,831
13 $60,000 to $74,999 3,466
14 $75,000 to $99,999 4,011
15 $100,000 to $149,999 3,706
16 $150,000 and over 2,635
Note. Complete cases only for all variables analyzed.
Materials and methods In GSOEP, participants were interviewed
face to face for the DRM questions and
through smartphones for the ESM
Samples
questions. In the DRM, as in the ATUS,
they were asked how they spent their time
We used data from ATUS wellbeing
yesterday and, for a random selection of
modules in 2010, 2012, and 2013. To
three activities, they were asked further
facilitate future replications of this
details about how they felt. In the ESM,
research, the ATUS extract builder was
participants were randomly notified on
used to create the dataset
mobile phones at seven random points
(https://www.atusdata.org) (Hofferth et al,
during the day for around one week. As in
2017). The ATUS is a repeated cross-
the DRM, they were asked how they were
sectional survey and is nationally
spending their time at the point of
representative of US household residents
notification, as well as how they felt.
aged 15 years and older. Its sampling
Participants in both ESM and DRM
frame is the Current Population Survey
samples were asked about whether they
(CPS), which was conducted two to five
were feeling happy, as well as other
months prior to the ATUS. Some items in
emotions such as sadness, stress and
the ATUS come from the CPS, including
boredom.
the household income item that we
analyze.
The focus of this research is on the
happiness items from both the ATUS and
Data from the GSOEP come from the
GSOEP to highlight differences according
Innovation Sample (IS), which is a
to the treatment of the independent
subsample of the larger main GSOEP
measure of income rather than differences
(Richter & Schupp, 2015). The main
according to the dependent outcome of
GSOEP and the IS are designed to be
emotional wellbeing.
nationally representative. The IS contains
information on household residents aged
Analyses
17 years of age and older. We used two
modules from these data: the 2012-2015
Data were analyzed in STATA 15. The
DRM module, which is a longitudinal
Supplementary Material S1 file contains
survey, and the 2014-15 ESM module.
the STATA ‘do’ file, which contains the
commands written to analyze the data. In
Measures
both ATUS and GSOEP, OLS regressions
were conducted with happiness as the
In ATUS, participants were called on the
outcome measure and income as the
phone and asked how they spent their time
explanatory measure. Following Kushlev
yesterday: what activities they were doing,
and colleagues (2015) and Hudson and
for how long, who they spent time with
colleagues (2016), the average happiness
and where they were located. This
across all activities each day was taken to
information was used to create their time
create an individual-level measure.
use diary. A random selection of three
Because the GSOEP DRM sample
activities were taken from these diaries
contained multiple observations across
and participants were asked how they felt
years, the standard errors were clustered at
during them. The feelings items were tired,
the individual level for models using this
sad, stressed, pain and happy. Participants
dataset.
were also asked how meaningful what they
were doing felt.
The treatment of income differed
according to the dataset because income
was collected differently in each dataset. also analyzed in continuous, log and
In the ATUS, income was first analyzed in quadratic forms.
continuous, log, and quadratic forms in
OLS regressions, as in other research We conducted lowess and spline
(Kushlev et al., 2015; Hudson et al., 2016). regressions to further investigate possible
Next, it was analyzed as a categorical non-linearities in the relationship between
variable with 16 categories, preserving the income and happiness. For the lowess
identical format that it was originally regressions, the bandwidth was set at the
collected in from the CPS questionnaire. default of 0.08. For the regression splines,
we fitted knots at four quartiles and five
In GSOEP, the income variable in the quantiles of income. We also used the
dataset is provided in continuous form results of OLS regressions treating income
because participants reported their as a categorical variable, as well as the
monthly income as an integer. To compare results of the lowess regression treating
to the ATUS results, 16 quantiles of income as continuous, to fit knots at pre-
income were created and analyzed in specified values of income where these
GSOEP DRMs (note that there were analyses suggested turning and/or satiation
insufficient observations to do this in points.
GSOEM ESMs). This income variable was

Table 2
The range and number of person-year observations of the GSOEP Income 4 variable
divided into 16 quantiles
Quanti
Income Income N
le
minimu maximu (observatio
numbe
m m ns)
r
1 2400 11520 433
2 11616 14400 459
3 14472 18000 584
4 18024 19200 228
5 19356 21600 427
6 21840 24000 520
7 24120 26880 306
8 26940 30000 660
9 30240 32400 257
10 33000 36000 631
11 36360 38400 193
12 39000 42000 430
13 42600 48000 539
14 49032 54000 289
15 54720 64800 400
16 66000 360000 410

Note. Complete cases only for all variables analyzed.

Controls
Results are presented without and with demographic characteristics and daily
controls for demographic and diary happiness (Kahneman et al., 2004; Stone
characteristics. Following Kushlev and et al., 2010; Luhmann et al., 2012: Hudson
colleagues (2015), Hudson and colleagues et al., 2019). These do not tend to be very
(2016), and Stone and colleagues (2018), large – for example, the effect size for the
these controls were age, gender, marital relationship between age and daily
status, ethnic backgroundi, healthii, experiences of happiness was 0.16 in
employment status, childreniii, and whether Stone et al. (2010). Our effect sizes range
the day was a weekend. We also control from 0.06 to 0.37.
for the year of the survey in DRM data to
address the issue that our results are not ATUS-DRM
due to new ATUS data but rather how we
treat the income variable. When treating the 16-category family
income variable as continuous in OLS
Robustness tests regressions, there was no relationship
(p>0.05) between income and happiness as
The results were assessed for consistency in other prior research (Kushlev et al.,
using multiple imputation and survey 2015; Hudson et al., 2015; Stone et al.,
weights. Multiple imputation by chained 2018). This was not observed when
equations (White et al, 2011) with 20 treating this variable as categorical: Our
iterations was used to assess bias due to results showed that those with incomes of
missing data (income and happiness in $35-40K were happier than those with
ATUS, 1.7% missing; income, happiness, higher incomes (see Figure 1). For
marital status, children, health, and example, with controls, those with
weekend in GSOEP DRMs, 8.2% missing; incomes of $35-40K were happier relative
income, age, marital status, employment, to those with incomes of $150K+ (b=0.16,
children, and health in ESM data, 6.0% se=0.04, p<0.001), $100-150K (b=0.14,
missing). Within the multiple imputation, se=0.04, p<0.001), and $50-60K (b=-0.08,
we applied person-level survey weights se=0.04, p=0.047). However, there were
(wbwt in ATUS and phrf in GSOEP). Any null results with multiple imputation and
differences to the main results are noted in survey weights, and results from
the text and Tables.iv regression splines and a lowess regression
also suggested null results overall (p>0.05;
see Figure 2). Further details of the
Results analyses are in S2.
In both ATUS and GSOEP, daily
happiness was analyzed using a 0-6 scale GSOEP-DRM
(in GSOEP scale points 1-7 were recoded
to 0-6 to match ATUS). The ATUS mean When treating the continuous household
happiness was 4.38 (sd=1.33). The income variable as continuous in OLS
GSOEP DRM mean happiness was 2.91 regressions, there was no relationship
(sd=1.46), and the GSOEP ESM mean (p>0.05) between income and happiness as
happiness was 2.65 (sd=1.03). in other prior research (Kushlev et al.,
2016; Hudson et al., 2016; Stone et al.,
Magnitude 2018). As in ATUS, this was not observed
when treating this variable as categorical:
There are not universally accepted Our results showed that those third
standards for evaluating effect sizes in quantile (~€14-18K) were happier than
happiness data. The magnitude of our those both higher and lower in income (see
results can be considered in the context of Figure 3). For example, they were happier
effect sizes from other research on
than those in quantiles 13 (€42.6K-48K,
b=0.46, se=0.14, p=0.001), seven (~€24-
27K, b=0.30, se=0.14, p=0.03), and one
(€2.40K-11,520, b=0.34, se=0.15, p=0.02).
However, the lowess and spline
regressions suggested null results overall,
as the variation was small in magnitude.
Further details of the analyses are in S3.

GSOEP-ESM

Summary of GSOEP ESM results

There was no evidence to suggest an

association between income and happiness
in ESM data for linear income, income
squared, log income, in the lowess
regressions, or regression splines (p>0.05).
A visualization of the lowess results are in
Figure 5 and further details of the analyses
are in S4.
Fig 1. Predicted values of average individual happiness in ATUS at the 16 values of the
family income variable without and with controls. Covariates at means. 95% confidence
interval. 4 .5
H a p p in e s s
4 .4
4 .3
4 .2

+
5K

t o 0K

K
20K

K
40K

K
60K

$10 t o K
K

to K
0K
5K

0K
<$

0K $30
$15

25

35

50

$75
7.5

00
K o<$1

15
12.

$15
$20 t o<$

<$

$35 t o<$

<$

$50 o<$

$60 o<$

0K <$1
<$

o<

<

o<

<$
t t t t
<$

to

to
to

t
t
5K K K K
.5K $15 $25 $40 K
K

K
K

K
$7.
K

$12 $75
$5

$3
$10

In c o m e
W ith o u t c o n tro ls W ith c o n tro ls
Fig 2. Line graph of predicted values from lowess regressions explaining variance in
happiness from income treated as a continuous variable in ATUS.
4.5

4.4

4.3

4.2
$5K .5K 10K .5K 15K 20K 25K 30K 35K 40K 50K 60K 75K 00K 50K 0K+
< 7 $ 2 $ $ $ $ $ $ $ $ $ 1 1 5
<$ o < <$1 o < o < o < o < o < o < o < o < o < <$ <$ $1
to t o t t t t t t t t t to to
5 K .5K K t .5K 5K 0K 5K 0K 5K 0K 0K 0K K K
$ 1 2 2 3 3 4 5 6 5 0
$7 $10 $12 $ $ $ $ $ $ $ $ $7 $10
Fig 3. Predicted values of average person-year happiness from GSOEP DRMs at 16
quantiles of income (Income 4) without and with controls. See Table 2 for income ranges in
each quantile. Covariates at means. 95% confidence interval.
Fig 4: Line graph of predicted values from lowess regressions explaining variance in
happiness from income treated as a continuous variable in GSOEP DRMs at 16
quantiles of income.

3.4
3.3
3.2
3.1
3
Happiness

2.9
2.8
2.7
2.6
2.5
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Income quantile (16)

Figure 5. Results of local linear ‘lowess’ regression from GSOEP ESM data with
happiness as the outcome and continuous annual income as the explanatory variable.
6
4
H a p p in e s s
2
0

0 € 20K € 40K € 60K € 80K € 100K

In c o m e

simply illustrates that a relationship

between income and happiness could be
Discussion interpreted when treating income
categorically rather than continuously,
The present results suggest that whether
which was justified in the Introduction as
income is associated with daily
being applicable to these data given the
experiences of happiness may depend on
arbitrary rates of change on the ATUS
how income is analyzed. When income in
income variable categories. Lowess and
ATUS is analyzed in its original,
spline regression analyses also suggested
categorical form, there is some evidence
that some people with higher incomes feel
that some people with higher incomes feel
less happy than those with lower incomes,
somewhat less happy than some of those
which is further evidence that the way
with lower incomes. When the continuous
income is analyzed matters.
income variable in GSOEP is split into
categories, a similar pattern is observed.
There are at least three possible
This is not inconsistent with the findings
interpretations to our overall results for the
of Kushlev and colleagues (2015) Hudson
effect of income on happiness. One
and colleagues (2016), and Stone and
interpretation tends towards conservative.
colleagues (2018), who found no
We conducted multiple comparisons of
relationship between income and daily
many transformations of income, which
feelings of happiness when income was
might inspire some to question whether we
analyzed as a continuous variable. It
should have accounted for this in some interpret these as suggesting the analytic
way by adjusting for multiple strategy for income could contribute to
comparisons. Although we found some explaining discrepant results in existing
evidence of differences in happiness literature, although sampling differences
according to income, such an adjustment could play a role, too.
might lead to an overall null conclusion
when characterising the effect of income Overall, the results were consistent
on happiness. A second interpretation is between countries in that there was a point
more generous. Within this perspective, in each dataset where higher income was
one might emphasise the fact that because not associated with higher happiness in
our income measures were correlated, no any model. In German DRM data,
correction for multiple comparisons was however, there was an untick in happiness
required. It could then be argued that after €52,800 that was (statistically)
because we found some evidence for significant when using survey weights –
effects of income on happiness, there is but this uptick did not surpass the level of
good evidence that the overall effect is not happiness associated with lower income
null. A more moderate perspective, and the levels. Despite a similar overall pattern in
one adopted in this paper, is that because the income results, there were other
the overall pattern of our results showed difference between countries. German
mixed null and non-null results, we can residents rated their happiness as lower
make an overall conclusion of some than US residents (a difference of ~1.5
effects of income on happiness. scale points out of seven). This could be
because of different interpretations of the
From the moderate perspective, these word ‘happiness’ in Germany and the
results both confirm and extend the United States. The word for happiness in
findings of Jebb and colleagues (2018), German used in the survey - glück – can
who found that higher income can be mean something more akin to lucky or
associated with worse evaluations of life. optimistic—which is different from the
Confirming the findings, we find some meaning of word ‘happy’ in the United
evidence that there can be satiation and/or States. Despite this linguistic difference,
turning points in the relationship between those with higher incomes were still less
income and happiness, whereby people happy than some of those with lower
with higher incomes are either no more or incomes in both samples.
less happy than some other lower income
groups. Extending the results, we show Limitations
that these points exist in how people feel
in the daily experience of their lives rather One limitation to our results is the
than only in overall evaluations of life. We representativeness of the income
also add to the conclusions of Kahneman distribution. Household surveys do not
and Deaton (2010), who found a satiation tend to capture the ‘tails’ of the income
point in measures of positive affect distribution very well: People in
yesterday in US data at ~$75K. Our data institutions and without addresses are
suggest that the satiation point is earlier excluded from these sample populations,
when DRM/ESM data are used: $35-40K which omits populations such as those
in the US and €14-18K in Germany. Our living in nursing homes and prisons, as
results do not support the results of Sacks well as the homeless. Moreover, people do
and colleagues (2012) or Killingsworth not always self-report their income
(2021), where no evidence of satiation accurately due to issues such as social
and/or turning points was found (see also desirability bias (Angel et al., 2019).
Stevenson and Wolfers, 2012). We Existing studies that have focused on those
with very low incomes do tend to find that and investigated for measures of emotional
low income is associated with low states other than happiness (Piff et al.,
happiness (Diener et al., 2002; Clark et al., 2018). In general, our results suggest that
2016; Adesanya et al., 2017). In ATUS, researchers should pay attention to how
the highest household income value income is measured and analyzed when
available was $150K, whereas in GSOEP considering how it is related to happiness,
it was €360K. Thus, it is not always clear which complements findings from other
whether the very affluent, such as research that the way happiness is
millionaires, are represented in these measured and analyzed is important
samples (Smeets, 2019). That survey (Kahneman & Deaton, 2010; Jebb et al.,
weights impacted the results in our 2018).
analyses suggest that this issue could
matter. Overall, our results cannot be taken Future research could also explore
as representative of people who are very mechanisms that may explain our findings.
poor or rich and should not be interpreted In addition to those mentioned in the
as such. Introduction – expectations, time use,
generosity and sense of self – another is the
Another limitation is that the present identity-related effect of transitioning
results cannot be interpreted casually between socio-economic groups. Though
because there has been no manipulation of one might expect upward mobility to be
income in these data nor exploration of associated with greater happiness, research
mechanisms. As discussed by Kushlev and suggests that some working class people do
colleagues (Kushlev et al., 2015), there are not wish to become upwardly mobile
issues such as reverse causality. Here, because it could lead to a loss of identity
however, our results potentially suggest an and change in community (Akerlof, 1997;
alternative reverse causality pathway, Friedman, 2014). Indeed, upward
whereby less happy people may select into intergenerational mobility is associated
earning more income. Because the with worse life evaluations in the United
counterfactual is not apparent – we do not Kingdom – though not in Switzerland
know how happy people with high (Hadjar & Samuel, 2015). Over time,
incomes would be without their higher therefore, the degree of mobility in a
income – it could also be that those with population could influence the relationship
high incomes would be even less happy between income and happiness in both
than they currently are if they had not positive and negative directions.
attained their current level of income. In
other words, people with high incomes Additionally, social comparisons could
may have started out as less happy in the drive the effects of higher income on
first place and be even less happy if they happiness. Higher income might not
didn’t have high incomes. benefit happiness if one’s reference group
– that is, the people to whom we compare
Future directions or have knowledge of in some form
(Hyman, 1942; Shibutani, 1955;
This research points to several directions Runciman, 1966) – changes with higher
for future research. One direction relates to socioeconomic status. As income increases,
data and measures: Non-linearities in the people might compare themselves to others
relationship between income and happiness who are also doing similarly or better to
could be examined using time use data them, and then not feel or think that they
from other countries, considered between are doing any better by comparison – or
countries and/or within countries over time even feel worse (Cheung & Lucas, 2016).
(Deaton et al., 2018; De Neve et al., 2018), It is, however, possible to view others’
greater success as one’s own future Akerlof, G A (1997). Social distance and
opportunity and for upward social social decisions. Econometrica:
comparisons to then positively impact upon Journal of the Econometric Society,
wellbeing (Senik, 2004; Davis & Wu, 1005-1027
2014; Ifcher et al., 2018)). As with the role Aknin, L B, Hamlin, J K, & Dunn, E W
of mobility in the relationship between (2012). Giving leads to happiness in
income and happiness, it is unclear whether young children.  PLoS One,  7(6).
the role of social comparisons would create Andreoni, J (1990). Impure altruism and
a positive or negative impact over time and donations to public goods: A theory
future research could explore this. of warm-glow giving. The Economic
Journal, 100(401); 464-477
Final remarks Angel, S, Disslbacher, F, Humer, S,
Schnetzer, M (2019). What did you
Overall, our results provide some evidence really earn last year? Explaining
that individual attainment in terms of measurement error in survey income
income may not equate to the attainment data. Journal of the Royal Statistical
of individual happiness – and could even Society: Series A (Statistics in
be associated with less daily happiness, Society), 182(4), 1411-1437.
depending upon how income is measured Balakrishnan, A, Palma, P A, Patenaude, J,
and analyzed. These results suggest that Campbell, L (2017). A 4-study
how income is associated with happiness replication of the moderating effects
depends on how income is measured and of greed on socioeconomic status
analyzed. They provide some support to and unethical behaviour. Scientific
the idea that financial achievement can Data, 4, 160120.
have both costs and benefits, potentially Bianchi, E C, Vohs, K D (2016). Social
informing normative discussions about the class and social worlds: Income
optimal distribution of income in society. predicts the frequency and nature of
social contact. Social Psychological
Acknowledgements and Personality Science, 7(5); 479-
486
Thank you to Professor Paul Dolan for his
Cheung, F, Lucas, R E (2016). Income
support in conducting this research.
inequality is associated with stronger
social comparison effects: The effect
Conflicts of interest of relative income on life
The authors have no conflicts of interest to satisfaction. Journal of Personality
report. and Social Psychology, 110(2); 332
Clark, A E, D’Ambrosio, C, Ghislandi, S
References (2016). Adaptation to poverty in
long-run panel data. Review of
Adesanya, O, Rojas, B M, Darboe, A,
Economics and Statistics. 98(3);
Beogo, I (2017). Socioeconomic
591-600.
differential in self-assessment of
Davis, L, Wu, S (2014). Social
health and happiness in 5 African
comparisons and life satisfaction
countries: Finding from World Value
across racial and ethnic groups: The
Survey. PloS One.:  12(11).
effects of status, information and
Aguiar, M, Hurst, E (2007). Measuring
solidarity. Social Indicators
trends in leisure: the allocation of
Research, 117(3); 849-869.
time over five decades. The
De Neve, J E, Ward, G, De Keulenaer, F,
Quarterly Journal of Economics,
Van Landeghem, B, Kavetsos, G,
122(3): 969-1006
Norton, M I (2018). The asymmetric
 experience of positive and negative Park, MD: University of Maryland
economic growth: Global evidence and Minneapolis, MN: University of
using subjective well-being data. Minnesota.
Review of Economics and Statistics, Hudson, N W, Lucas, R E, & Donnellan,
100(2); 362-375 M B. (2019). Healthier and happier?
Deaton, A (2008). Income, health, and A 3-year longitudinal investigation
well-being around the world: of the prospective associations and
Evidence from the Gallup World concurrent changes in health and
Poll. Journal of Economic experiential well-being. Personality
Perspectives, 22(2); 53-72. and Social Psychology
Diener, E, Biswas-Diener, R (2002). Will Bulletin, 45(12); 1635-1650.
money increase subjective well- Hudson, N W, Lucas, R E, Donnellan, M
being? Social Indicators B, & Kushlev, K (2016). Income
Research,  57(2), 119-169. reliably predicts daily sadness, but
Dolan, P, Peasgood, T, White, M (2008). not happiness: A replication and
Do we realy know what makes us extension of Kushlev, Dunn, and
happy? A review of the economic Lucas (2015). Social Psychological
literaure on the factors associated and Personality Science, 7(8); 828-
with subjective well-being. Journal 836.
of Economic Psychology: 29(1); 94– Hyman, H H. The psychology of status
122 . (1942). Archives of Psychology
Dunn, E W, Aknin, L B, & Norton, M I (Columbia University),
(2008). Spending money on others Ifcher, J, Zarghamee, H, Graham, C.
promotes happiness. Science, (2018) Local neighbors as positives,
319(5870); 1687-1688 regional neighbors as negatives:
Friedman, S (2014). The price of the Competing channels in the
ticket: Rethinking the experience of relationship between others’ income,
social mobility. Sociology, 48(2); health, and happiness. Journal of
352-368. Health Economics,  57, 263-276
Graham C, Pettinato, S (2002). Frustrated Jebb, A T, Tay, L, Diener, E, & Oishi, S
achievers: Winners, losers and (2018). Happiness, income satiation
subjective well-being in new market and turning points around the
economies. Journal of Development world. Nature Human Behaviour,
Studies,  38(4): 100-140. 2(1); 33.
Hadjar, A, Samuel, R (2015). Does Kahneman, D, & Deaton, A (2010). High
upward social mobility increase life income improves evaluation of life
satisfaction? A longitudinal analysis but not emotional well-being.
using British and Swiss panel Proceedings of the National
data. Research in Social Academy of Sciences, 107(38),
Stratification and Mobility, 39; 48- 16489–16493.
58. Kahneman, D, Krueger, A, Schkade, D,
Hamermesh, D S, Lee, J (2007). Stressed Schwarz, N, Stone, A (2004). A
out on four continents: Time crunch survey method for characterizing
or yuppie kvetch? The Review of daily life experience: the day
Economics and Statistics, 89(2): reconstruction method. Science,
374-383 306(5702), 1776–1780.
Hofferth, S, Flood, S, Sobek M (2017). Killingsworth, M. A. (2021). Experienced
American Time Use Survey Data well-being rises with income, even
Extract System: Version 26 above $75,000 per year.
[Machine-readable database] College Proceedings of the National
 Academy of Sciences, 118(4). Runciman, W (1966). Relative
Kraus, M W, Côté, S, Keltner, D (2010). Deprivation, Social Justice: Study
Social class, contextualism, and Attitudes Social Inequality in 20th
empathic accuracy. Psychological Century England University of
Science,  21(11); 1716-1723 California Press.
Kushlev, K, Dunn, E W, Lucas, R E Sacks, D W, Stevenson, B, Wolfers, J
(2015). Higher income is associated (2012). The new stylized facts about
with less daily sadness but not more income and subjective well-being.
daily happiness. Social Emotion, 12(6), 1181.
Psychological and Personality Senik, C. (2004). When information
Science, 6(5); 483–489. dominates comparison: Learning
Luhmann, M, Hofmann, W, Eid, M, from Russian subjective panel data.
Lucas, R E (2012). Subjective well- Journal of Public Economics, 88(9-
being and adaptation to life events: a 10), 2099-2123.
meta-analysis. Journal of Shibutani, T (1955). Reference groups as
Personality and Social Psychology, perspectives.  American Journal of
102(3), 592. Sociology. 1055: 60(6); 562-569.
Nickerson C, Schwarz, N, Diener, E, Smeets, P., Whillans, A., Bekkers, R., &
Kahneman, D (2003). Zeroing in on Norton, M. I. (2020). Time use and
the dark side of the American dream: happiness of millionaires: Evidence
A closer look at the negative from the Netherlands. Social
consequences of the goal for Psychological and Personality
financial success. Psychological Science, 11(3), 295-307.
Science,  14(6): 531-536 Snibbe, A C, & Markus, H R (2005). You
Nikolaev, B (2018). Does Higher can't always get what you want:
Education Increase Hedonic and educational attainment, agency, and
Eudaimonic Happiness? Journal of choice. Journal of Personality and
Happiness Studies, 19(2): 483-504 Social Psychology, 88(4); 703.
Piff, P K, Kraus, M W, Côté, S, Cheng, B Stephens, N M, Markus, H R, &
H, & Keltner, D (2010). Having less, Townsend, S (2007). Choice as an
giving more: the influence of social act of meaning: the case of social
class on prosocial behavior. Journal class.  Journal of Personality and
of Personality and Social Social Psychology, 93(5); 814.
Psychology, 99(5); 771. Stevenson, B, Wolfers, J (2012).
Piff, P K, Moskowitz, J P (2018). Wealth, Subjective well-being and income: Is
poverty, and happiness: Social class there any evidence of
is differentially associated with satiation? American Economic
positive emotions. Emotion, 18(6); Review, 103(3); 598-604.
902. Stone, A A, Schwartz, J E, Broderick, J E,
Richter, D, Schupp, J (2015). The SOEP Deaton, A. (2010). A snapshot of the
Innovation Sample (SOEP IS) age distribution of psychological
Schmollers Jahrbuch: 135 (3); 389- well-being in the United
399 (doi: 103790/schm1353389)  States. Proceedings of the National
(prepublished 2012: SOEP Academy of Sciences, 107(22); 9985-
Innovation Sample (SOEP-IS) - 9990.
Description, Structure and Stone, A, Schneider, S, Krueger, A,
Documentation SOEPpaper 463 | Schwartz, J E, & Deaton, A (2018).
PDF, 16291 KB : Berlin DIW Experiential wellbeing data from the
Berlin) American Time Use Survey:
comparisons with other methods and
 analytic illustrations with age and
income. Social Indicators Research,
136(1); 359-378.
White, I R, Royston, P, Wood, A M
(2011). Multiple imputation using
chained equations: issues and
guidance for practice. Statistics in
Medicine, 30(4); 377–399.
Supporting information S2 Files. ATUS DRM results
S3 Files. GSOEP DRM results
S1 File. STATA do file.
S4 Files. GSOEP ESM results
S1 File. STATA do file
i
In the ATUS this was Hispanic and Black, in GSOEP this was German origin.
ii
In the ATUS this was whether the respondent had any physical or cognitive difficulty (yes/no), in

GSOEP this was self-rated general health (bad, poor, satisfactory, good, very good).
iii
In the ATUS this was presence of children < 18 years in the household, in GSOEP this was

number of children.
iv
Note that the lowess analyses were conducted without controls, survey weights or multiple

imputation due to the nature of this estimation method.