R ES E A RC H

◥ RESULTS: The data revealed substantial cross-

RESEARCH ARTICLE SUMMARY country variation in gender differences in pref-
erences. Gender differences were found to be
strongly positively associated with economic
SOCIAL SCIENCE
development as well as gender equality. These
relationships held for each preference separately
Relationship of gender differences in as well as for a summary index of differences
in all preferences jointly. Quantitatively, this

preferences to economic development ON OUR WEBSITE

◥
summary index exhibited
correlations of 0.67 (P <
0.0001) with log GDP per
and gender equality Read the full article
at http://dx.doi.
org/10.1126/
capita and 0.56 (P < 0.0001)
with a Gender Equality
science.aas9899 Index (a joint measure
Armin Falk* and Johannes Hermle* ..................................................
of four indices of gender
equality), respectively. To isolate the separate
INTRODUCTION: Understanding determi- development and gender equality (social role impacts of economic development and gender
nants of gender differences in economic and hypothesis). On the other hand, greater avail- equality, we conducted a conditional analysis,
social domains has been of interest, both in ac- ability of material and social resources removes finding a quantitatively large and statistically
ademic and public debates. Previous research the gender-neutral goal of subsistence, which significant association between gender differ-

Downloaded from http://science.sciencemag.org/ on November 6, 2018

has shown that gender differences in funda- creates the scope for gender-specific ambitions ences and log GDP per capita conditional on the
mental economic preferences are important and desires. In addition, more gender-equal Gender Equality Index, and vice versa. These
in explaining gender differences in economic access to those resources may allow women findings remained robust in several validation
outcomes, such as for occupational choice, fi- and men to express preferences independently tests, such as accounting for potential culture-
nancial investment, or educational decisions, from each other. As a consequence, one would specific survey response behavior, aggregation
among many others. However, gaps remain expect gender differences in preferences to bias, and nonlinear relationships.
in understanding the sources of gender dif- be positively associated with higher levels of
CONCLUSION: The reported evidence indicates
ferences in preferences and their variation. economic development and gender equality
that higher levels of economic development and
(resource hypothesis).
gender equality favor the manifestation of gen-
RATIONALE: We contrasted and tested two We tested these competing predictions using
der differences in preferences across countries.
hypotheses that make opposite predictions data on experimentally validated measures of
Our results highlight the critical role of avail-
concerning the cross-country association of willingness to take risks, patience, altruism,
ability of material and social resources, as well
gender differences in preferences with eco- positive and negative reciprocity, and trust
as gender-equal access to these resources, in
nomic development and gender equality. On for 80,000 individuals in 76 representative
facilitating the independent formation and ex-
one hand, the attenuation of gender-specific
social roles that arises in more developed and
country samples. So that the data would be
geographically representative, the dataset was
pression of gender-specific preferences.
▪
gender-egalitarian countries may alleviate dif- chosen so as to include all continents and a
ferences in preferences between women and broad range of cultures and economic devel- The list of author affiliations is available in the full article online.
men. As a consequence, one would expect gen- opment levels. In total, the data represent *Corresponding author. Email: [email protected]
(A.F.); [email protected] (J.H.)
der differences in preferences to be negative- about 90% of both the world population and Cite this article as A. Falk, J. Hermle, Science 362, eaas9899
ly associated with higher levels of economic global income. (2018). DOI: 10.1126/science.aas9899

1 Economic Development
.8
.6
.4
Average Gender Difference

.2
in Preferences (Index)

0
6 7 8 9 10
Log GDP p/c
1 Gender Equality
.8
.6
Gender Differences in Preferences .4
.2
Larger Smaller
0
No Data 0 .2 .4 .6 .8 1
Gender Equality Index

Gender differences in preferences across countries and their association with economic development and gender equality. (Left)
World map visualizing a summary index of gender differences in all six preferences (risk-taking, patience, altruism, trust, and positive and
negative reciprocity). (Right) The relationship between the summary index of gender differences in preferences and (top) log GDP per capita
and (bottom) a Gender Equality Index comprising measures of material, social, and political gender equality. The relationships are predicted
from local polynomial regressions. Shaded areas indicate 95% confidence intervals.

Falk et al., Science 362, 307 (2018) 19 October 2018 1 of 1

R ES E A RC H

◥ Data and measures

RESEARCH ARTICLE An empirical test of the two competing hypothe-
ses requires data that meet three critical condi-
tions: (i) reliability of preference measures, (ii)
SOCIAL SCIENCE extensive cultural variation as well as compre-
hensive global coverage, and (iii) representative-

Relationship of gender differences in ness of country samples. Our investigation used

the Global Preference Survey (GPS) (22, 23). The
GPS was collected as part of the Gallup World
preferences to economic development Poll 2012 and contains measures of six funda-
mental preferences with regard to social and

and gender equality nonsocial domains: willingness to take risks;

patience, which captures preferences over the
intertemporal timing of rewards; altruism; trust
Armin Falk1* and Johannes Hermle2* (24); and positive and negative reciprocity, which
capture the costly willingness to reward kind
Preferences concerning time, risk, and social interactions systematically shape human actions or to punish unkind actions, respectively.
behavior and contribute to differential economic and social outcomes between women Before the launch of the international survey,
and men. We present a global investigation of gender differences in six fundamental multiple survey items were selected for these
preferences. Our data consist of measures of willingness to take risks, patience, altruism, preferences through an ex ante experimental
positive and negative reciprocity, and trust for 80,000 individuals in 76 representative validation (25). For each preference, subjects

Downloaded from http://science.sciencemag.org/ on November 6, 2018

country samples. Gender differences in preferences were positively related to economic responded to a large set of survey items and
development and gender equality. This finding suggests that greater availability of participated in incentivized choice experiments.
and gender-equal access to material and social resources favor the manifestation of The subset of survey items that maximized ad-
gender-differentiated preferences across countries. justed R2 in predicting incentivized behavior in

F
the corresponding experiment was selected for
undamental preferences such as altruism, nent in explaining the gender preference gap. the international survey. The selected items,
risk-taking, reciprocity, patience, or trust As a consequence, according to the social role described below, comprise a combination of
constitute the foundation of choice theories hypothesis, higher economic development and qualitative self-assessments and quantitative
and govern human behavior. A growing lit- gender equality (and the associated dissolution items that involve economic trade-off decisions.
erature in economics (1, 2) and psychology of traditional gender roles) should lead to a nar- The qualitative items elicit participants’ subjec-
(3) documents important gender differences in rowing of gender differences in preferences. tive assessment of their willingness to act in a
preferences. These differences provide a key ex- In contrast, there is reason to expect that gen- certain way, such as whether participants are
planation for differential choices and outcomes der differences in preferences expand with eco- generally willing to take risks. Complementarily,
between women and men in contexts such as nomic development and gender equality (resource the quantitative items provide revealed prefer-
occupational choice, financial investment, or hypothesis). As suggested by post-materialist ence measures by using participants’ choices in
educational decisions (4, 5), among many others. theory (15, 16), a critical societal precondition for monetary trade-off decisions. As an example,
The origins of gender differences in preferences, self-expression is the fulfillment of basic material the quantitative item for risk-taking provides
and their variability across countries and cultures, needs. In line with this, existing research shows the participants with a sequence of five inter-
are addressed by an extensive literature that dis- that the unrestricted expression of preferences dependent choices between a fixed and a risky
usses biological and evolutionary determinants hinges on the availability of sufficient material payment (lottery). This allows one to progres-
(6, 7) and the role of the social environment (8–10). and social resources (17–20). Therefore, gender sively approach the point of indifference between
differences in preferences should manifest them- the fixed payment and the lottery, which serves
Hypotheses selves only if both women and men obtain suffi- as a revealed preference measure for risk-taking
We contrast two competing hypotheses that make cient access to these resources to independently behavior. The presence of both qualitative and
opposite predictions concerning the cross-country develop and express their intrinsic preferences quantitative items allows for robustness tests
correlational patterns of gender differences in (21). Specifically, greater availability of material with respect to potential culture-specific response
preferences with respect to economic develop- resources eliminates the gender-neutral goal of behavior. So that survey items were comparable
ment and gender equality. Following social role subsistence. This creates scope for attending to across cultures, all items were translated back
theory, one may hypothesize that gender differ- gender-specific ambitions and desires. As a con- and forth by professionals, and monetary values
ences in preferences attenuate in more developed, sequence, economic development may facilitate mentioned in the survey questions were adjusted
gender-egalitarian countries (social role hypoth- the unfolding of differences between women along median household income across countries.
esis). This hypothesis rests on two premises. and men. More developed countries also feature The survey items were pretested in 22 countries
First, economic development is a key determinant higher levels of gender equality in political, social, of varied cultural heritage as part of the Gallup
of societal progression toward gender equality and economic domains (11), which is a critical re- World Poll 2012 pretest, conducted in late 2011,
(11, 12), which is critical for the dissolution of quirement for the acceptance of gender-specific to guarantee cross-cultural validity.
traditional gender roles (13, 14). Second, as dis- desires and preferences. In particular, as women After the ex ante experimental validation and
cussed by a large body of literature (8–10), gender- become less exposed and vulnerable to male in- pretests, the international survey was imple-
specific roles instill distinct preferences in women fluence, gender differentiation may be reinforced mented in a total of 76 countries, representing
and men and hence constitute a crucial compo- through women’s greater opportunities for self- about 90% of the global population and global
expression. In sum, greater availability of mate- GDP. To provide geographic representativeness
rial and social resources to both women and men as well as developmental and cultural variation,
1
Department of Economics, University of Bonn, and Behavior may facilitate the independent development and we selected the countries to include all conti-
and Inequality Research Institute (briq), 53113 Bonn, Germany. expression of gender-specific preferences, and nents and a very broad range of economic devel-
2
Department of Economics, University of California, Berkeley,
CA 94720, USA.
hence may lead to an expansion of gender differ- opment levels. For each country, the data contain
*Corresponding author. Email: [email protected] (A.F.); ences in more developed and gender-egalitarian samples representative of the resident popula-
[email protected] (J.H.) countries. tion aged 15 and older, with a median sample size

Falk et al., Science 362, eaas9899 (2018) 19 October 2018 1 of 6

R ES E A RC H | R E S EA R C H A R T I C LE

A Altruism (+) Trust (+) Pos. Recip. (+) B Correlation: 0.6685 (p<0.0001)
Average Gender Difference (in Standard Deviations)
.25

.25

.25
CAN

1
USA
GBR
SWE
.2

.2

.2
AUS
NLD
AUT
.15

.15

.15

Average Gender Difference (Index)

ESTCZE FIN CHE

.8
.1

.1

.1
.05

.05

.05
BOL
RUS
UKR DZA ARG LTU JPN
0

0
DEU
POLHUN

.6
−.05

−.05

−.05
ZWE SAU FRA
Poorer Countries Richer Countries Poorer Countries Richer Countries Poorer Countries Richer Countries CHN BWA ISR
MAR PRT ESP
EGY COL
Neg. Recip. (−) Risk taking (−) Patience (−) UGA IND SUR
SRB CHLHRV
NGA GEO GRC ITA
KOR
.25

.25

.4
IDN PER
.25

ROU
RWA PHL CRIVENTUR
HTI JOR
.2

.2
.2

MWI BRA
MDA GTMBIH KAZ ARE
CMR ZAF
.15

.15
.15

VNM MEX

.2
AFG KEN
.1

.1
LKA
.1

BGD NIC THA

.05

.05
.05

KHM
IRN
0

0
0

TZA GHAPAK IRQ

0
−.05

−.05

−.05
Poorer Countries Richer Countries Poorer Countries Richer Countries Poorer Countries Richer Countries 6 7 8 9 10 11
Log GDP p/c

C Altruism (+) Trust (+) Pos. Recip. (+) D Correlation: 0.5580 (p<0.0001)
CAN

1
Average Gender Difference (in Standard Deviations)

Downloaded from http://science.sciencemag.org/ on November 6, 2018

.25

.25

.25

USA
GBR
SWE
.2
.2

.2

AUSNLD
AUT
.15
.15

.15

Average Gender Difference (Index)

CHE EST
CZE FIN

.8
.1
.1

.1
.05
.05

.05

BOL
RUS LTU
DZA ARG
JPNUKR
−.05 0
−.05 0

−.05 0

DEU
.6 HUN POL
SAU ZWE FRA
Less Gender− More Gender− Less Gender− More Gender− Less Gender− More Gender− BWA CHN ISR ESP
Equal Countries Equal Countries Equal Countries Equal Countries Equal Countries Equal Countries MAR PRT
EGY SUR COL HRV
Neg. Recip. (−) Risk taking (−) Patience (−) IND CHL
UGA SRB
IDN PERKORGRCGEOITA
.4
.25

.25

ROU
.25

TUR VEN CRI PHL RWA

JOR BRA
MWI
.2

.2
.2

ARE GTM MDA KAZ

CMR ZAF
.15

.15
.15

MEX VNM
.2

KEN
BGD LKA NIC THA
.1

.1
.1
.05

.05
.05

KHM
IRN
−.05 0

−.05 0
−.05 0

PAK TZA
GHA
0

Less Gender− More Gender− Less Gender− More Gender− Less Gender− More Gender−
Equal Countries Equal Countries Equal Countries Equal Countries Equal Countries Equal Countries 0 .2 .4 .6 .8 1
Gender Equality (Index)

Fig. 1. Analysis of gender differences in preferences in relation to correspond to women exhibiting higher levels of the respective
economic development and gender equality. (A) Mean country-level preference, whereas − indicates that positive differences correspond
gender difference in altruism, trust, positive reciprocity, negative to women exhibiting lower levels of the respective preference.
reciprocity, risk-taking, and patience by development level. Countries (B) Relationship between the aggregate index of gender differences
were sorted into four bins according to their GDP per capita quartile. in all six preferences and log GDP per capita. (C and D) Same
The symbols + and − in the panel titles indicate the sign of the relationships as in (A) and (B) for the Gender Equality Index.
difference for each preference; + indicates that positive differences See supplementary materials for country abbreviation key.

of 1000 participants per country; this made gen- of 0 and a standard deviation of 1. Then, for each and less patient by 0.050 standard deviations (P <
eralizable inferences possible. In total, the data preference, an ordinary least-squares (OLS) re- 0.0001) (26). The observed differences in prefer-
include preference measures for about 80,000 gression was performed on the worldwide sam- ences set the stage for our analysis.
participants. ple, using as the independent variable a gender
After implementation of the worldwide survey, indicator in which male is the reference category, Analysis of gender differences in
the measures for the six preferences were gen- controlling for age, age squared, cognitive skills preferences in relation to economic
erated according to the following procedure. (as proxied by subjective math skills), education development and gender equality
First, each of the survey items was standardized level, household income quintile, and country To test the competing hypotheses, we computed
using the mean and variance of the entire world- fixed effects. Standard errors were clustered at country-level gender differences for each pref-
wide sample. Then, to obtain the preference mea- the country level. The estimated coefficient on the erence. For this purpose, we standardized each
sures, the relevant z-scores were averaged using gender indicator served as the gender difference preference measure at the country level to exhibit
weights developed in the experimental valida- in the respective preference. On the global level, a mean of 0 and a standard deviation of 1. We
tion. Further details on the data collection and all six preferences featured significant gender then performed for each preference and country
construction of our measures are given below and differences (fig. S1): Women tended to be more a separate OLS regression using as independent
in the supplementary materials. prosocial and less negatively reciprocal than men, variable a gender indicator in which male is the
The data allow assessment of the existence with differences in standard deviations of 0.106 for reference category. We also included several con-
and quantitative relevance of gender differences altruism (P < 0.0001), 0.064 for trust (P < 0.0001), trols to isolate the gender effect from potentially
in preferences at the global level (22). For this 0.055 for positive reciprocity (P < 0.0001), and confounding factors that might differ between
purpose, global gender differences were calcu- 0.129 for negative reciprocity (P < 0.0001). Turning women and men. These controls are age, age
lated as follows: Each preference measure was to nonsocial preferences, women were less risk- squared, subjective math skills, education level,
standardized at the global level to exhibit a mean taking by 0.168 standard deviations (P < 0.0001) and household income quintile. The obtained

Falk et al., Science 362, eaas9899 (2018) 19 October 2018 2 of 6

R ES E A RC H | R E S EA R C H A R T I C LE

coefficient on the gender indicator served as a we conducted a conditional analysis. We con- quantitative items (fig. S7, A and B). A test of
measure of the gender difference in the respec- structed partial regression plots illustrating the the null hypothesis of equality of the correlation
tive preference and country. relationship between the index of gender differ- coefficients failed to reject at conventional sig-
Using the country-level estimates of gender ences in preferences and log GDP per capita con- nificance levels (P = 0.744). Likewise, correla-
differences in preferences, we examined varia- ditional on the Gender Equality Index (Fig. 2A) tions with the Gender Equality Index were 0.480
tion along levels of economic development and and vice versa (Fig. 2B). The dependent and inde- (P < 0.0001) for qualitative items and 0.479 (P <
gender equality. As the measure of economic de- pendent variables were standardized to exhibit a 0.0001) for quantitative items (fig. S7, C and D).
velopment, we used GDP per capita. To assess mean of 0 and a standard deviation of 1. Hence, Testing equality of the coefficients failed to re-
the role of gender equality, we created a Gender the slope coefficients can be interpreted as the ject (P = 0.991), thus providing no support that
Equality Index as a joint measure of four indices standard deviation change in the dependent var- culture-specific response behavior contaminated
of gender equality: (i) the Global Gender Gap iable in response to a change of one standard the results.
Index of the World Economic Forum (WEF), (ii) deviation in the independent variable. To further test for the robustness of our re-
the Gender Equality Index of the United Nations There was a quantitatively large and statis- sults, we conducted several additional analyses.
(UN), (iii) the ratio of female to male labor force tically significant association of gender differ- First, because trust reflects a composite trait that
participation rates, and (iv) the number of years ences with log GDP per capita conditional on the captures beliefs about others’ behavior, prosocial
since women’s suffrage. The Gender Equality Gender Equality Index. The estimated slope co- preferences, and preferences for risk-taking, we
Index was constructed as the predicted main efficient was 0.53 (P < 0.0001). Likewise, gender repeated our analysis excluding the trust dimen-
component from a principal components analy- differences were strongly associated with the sion. To do so, we constructed an alternative index
sis of the four indices. Gender Equality Index conditional on log GDP of gender differences in preferences in a proce-
To study the effect of economic development, per capita, with a somewhat smaller slope co- dure parallel to the main index but using only the
we first sorted the 76 countries into four bins efficient of 0.32 (P = 0.0033) (see also table S4, five remaining preferences. Similar to our main

Downloaded from http://science.sciencemag.org/ on November 6, 2018

according to their level of development, measured column 7). When conducting an F-test for equal- results, this alternative index exhibited a quanti-
by GDP per capita. We then computed for each ity of both coefficients, we failed to reject at P = tatively large association with economic develop-
bin the average country-level gender difference 0.2537, indicating that the strength of the rela- ment and measures of gender equality (tables S5
in each preference. Gender differences in all six tionships between the index of gender differ- and S6). Second, we tested whether the level of
preferences increased with a country’s level of ences in preferences and log GDP per capita and standardization affected our results. We repeated
development (Fig. 1A). The positive correlations the Gender Equality Index were not statisti- our analysis using preference measures stan-
between log GDP per capita and country-level cally different. These findings imply that both dardized at the global rather than the country
gender differences were large and statistically economic development and gender equality ex- level. The results using preferences standardized
significant for all six preferences: 0.58 for altru- hibited an independent and significant associ- at the global level were similar to our main re-
ism (P < 0.0001), 0.59 for trust (P < 0.0001), 0.31 ation with gender differences in preferences (28). sults (fig. S8 and tables S7 and S8). Third, we
for positive reciprocity (P = 0.0067), 0.35 for Conditional on log GDP per capita, differences repeated our analysis without using individual-
negative reciprocity (P = 0.0017), 0.37 for risk- in preferences were also significantly and posi- level controls when calculating gender differ-
taking (P = 0.0011), and 0.38 for patience (P = tively associated with the four individual mea- ences, yielding similar results (fig. S9 and tables
0.0006) (fig. S2). We also analyzed a summary sures of gender equality (Fig. 2, C to F). Slope S9 and S10). Fourth, a common concern in cross-
index of gender differences for all preferences coefficients were 0.23 (P = 0.0084) for the WEF country analysis involves measurement error. Be-
jointly. For this purpose, we first performed a Global Gender Gap Index, 0.29 (P = 0.0515) for cause the experimental validation was conducted
principal components analysis of the country- the UN Gender Equality Index, 0.25 (P = 0.0123) in Germany, more linguistically similar countries
level gender differences in the six preferences. for ratio of female to male labor force partici- might exhibit smaller measurement error. To test
We then created an index of gender differences pation, and 0.30 (P = 0.0023) for years since for robustness against this potential confound,
in preferences as the predicted main component. women’s suffrage. we additionally controlled for linguistic distance
This index exhibited a correlation of 0.67 (P < In sum, these findings provide evidence in favor to German, which left the results qualitatively
0.0001) with log GDP per capita (Fig. 1B) (27). of the resource hypothesis that higher levels of unchanged (tables S11 and S12). Fifth, to address
To study the effect of gender equality, we ran economic development and gender quality are concerns of aggregation bias, we tested for the
the same analysis as for economic development, associated with stronger gender differentiation relationship between household income and
using the Gender Equality Index as the explan- in preferences. gender differences in preferences in individual-
atory variable. Gender differences in preferences A potential concern regarding the reported level regressions, finding a significant relation-
were found to increase with gender equality for results involves bias due to culture-specific sur- ship for each preference (table S13). Finally, we
each preference separately (Fig. 1C) as well as vey response behavior (29–32). Note that our tested for a nonlinear relationship with eco-
for the index of gender differences in preferences data contain two types of items: qualitative self- nomic development. A closer inspection of Fig.
(Fig. 1D). For the individual preferences, the cor- assessments and quantitative choice measures. 1B suggested a nonlinear, convex relationship,
relation coefficients were 0.51 for altruism (P < Qualitative self-assessments might be affected which is confirmed by regression analysis (table
0.0001), 0.41 for trust (P = 0.0005), 0.13 for pos- by response biases such as scaling effects, which S14, column 2). This pattern originated from the
itive reciprocity (P = 0.2875), 0.40 for negative might vary across cultures, thereby introducing fact that richer countries are overproportionally
reciprocity (P = 0.0005), 0.34 for risk-taking (P = systematic measurement error (33). In contrast, more gender-equal. Therefore, when we inves-
0.0036), and 0.43 for patience (P = 0.0002) (fig. the quantitative items present trade-offs that are tigated the relationship between the index of
S3). The summary index of gender differences in well-defined in terms of stakes and probabilities, gender differences in preferences and log GDP
preferences exhibited a correlation of 0.56 (P < yielding revealed preference measures that facil- per capita after residualizing both variables with
0.0001) with the Gender Equality Index. Reassur- itate a culturally fair comparison. To test for ro- respect to the Gender Equality Index, the rela-
ingly, the positive relationship between the index bustness with regard to the elicitation method, tionship was found to be linear (table S14). See
of gender differences in preferences and gender we constructed two separate indices of gender supplementary text for details of the robust-
equality was also found for the four individual differences using either qualitative or quantita- ness tests.
indicators of gender equality (fig. S4). tive items only (analogous to how we constructed
Economic development and gender equality the main index). The correlations of the indices Concluding remarks
are strongly intertwined (11). To isolate the sepa- with log GDP per capita were found to be very The reported evidence indicates that higher levels
rate impacts of economic development and gen- similar, with values of 0.551 (P < 0.0001) for of economic development and gender equality
der equality on gender differences in preferences, qualitative items and 0.516 (P < 0.0001) for are associated with stronger gender differentiation

Falk et al., Science 362, eaas9899 (2018) 19 October 2018 3 of 6

R ES E A RC H | R E S EA R C H A R T I C LE

in preferences. These findings may also relate also do not preclude a role for biological or evo- In this regard, our findings point toward the
to other personality traits, such as the Big Five lutionary determinants of gender differences critical role of availability of and equal access
(34, 35) or value priorities (36). Our findings do (37). Our results highlight, however, that theo- to material and social resources for both women
not rule out an influence of gender-specific roles ries not attributing a significant role to the social and men in facilitating the independent forma-
that drive gender differences in preferences. They environment are incomplete (38). tion and expression of gender-specific preferences

A Economic Development B Gender Equality

(Residualized using Gender Equality Index)
2

2
DZA SAU
Average Gender Difference (Index)

Average Gender Difference (Index)

USA BOL

(Residualized using Log GDP p/c)

CAN
MAR GBR ZWE
CAN
EGY
CZE AUS AUT CHE USA GBR UKR
EST
1

1
NLD DZA CZEUGAAUS SWE
BOL ARG
EST SWE
JOR JPN ARG AUT NLD
RUS
CHN RWA
IND MWI LTU
IND RUS MAR
EGY
ZWE UKR BWA FIN
SURHUN CHE POL
CHN LTUPOL FIN IDN
BWA GEO
IDN COL ARE COL HUN SRBMDA
TUR SUR PHL
0

0
UGA PER CHL ISR SAU JPN PER CMR
GTM VEN DEU KOR
FRA
ESP GRC BGD KEN VNM DEU
SRB HRV JOR
GEO CRI PRT IRN GTM CHL ISR ROU PRT
HRV
MWI CMR ROUBRA ITA CRI
VEN BRA FRA
ESP
KHM
PAK TUR LKA NIC KAZ
PHL
BGD MEX
RWA
MDA KOR
GRC ZAFTZA
−1

−1
KEN LKA PAK THA
GHA
VNM NIC KAZ ZAF ITA
KHM THA MEX
IRN
ARE
TZA GHA Slope Coefficient: 0.5258 (p<0.0001) Slope Coefficient: 0.3192 (p=0.003)

Downloaded from http://science.sciencemag.org/ on November 6, 2018

−2

−2
−2 −1 0 1 2 −3 −2 −1 0 1 2
Log GDP p/c Gender Equality (Index)
(Residualized using Gender Equality Index) (Residualized using Log GDP p/c)

C WEF Global Gender Gap Index D UN Gender Equality Index

2
2

BOL BOL
Average Gender Difference (Index)
Average Gender Difference (Index)

(Residualized using Log GDP p/c)

(Residualized using Log GDP p/c)

ZWE ZWE
CAN CAN
UKR
USA GBR USA UKR
EST GBR UGA EST
1
1

DZA CZE UGA SWE DZA AUS SWE CZE

AUS NLD
RUS AUT ARG NLD ARG RUS AUT
INDCHN MWI RWA IND MWI RWACHN
MAR
EGY LTU MAR
EGY LTU
NGA FIN HTI FIN
POL CHE CHE POL
IDN
GEOBWA BWA IDN GEO
HUN COL AFGCOL HUN
SUR MDA PHL SUR PHL SRB MDA
0

SRB
0

SAU JPN PERCMR CMR SAU PER JPN VNM

VNM BGDKEN DEU KEN BGD DEU
JOR CHL
ISR JORCHL ISR
GTM ROU HRV PRT GTM ROU PRTHRV
KHM
VEN CRIFRA
ESP VEN CRI FRAESP
KHM BIH
TUR BRA LKA KAZ NIC BRA TUR NIC KAZ
LKA
TZA ZAF ZAF TZA
KOR GRC GRC KOR
−1
−1

PAK THA GHA PAKTHA

GHA
ITA ITA
MEX MEX
IRN IRN
IRQ
ARE ARE
Slope coefficient: 0.2327 (p=0.008) Slope coefficient: 0.2911 (p=0.052)
−2
−2

−2 −1 0 1 2 −1 −.5 0 .5 1 1.5
WEF Global Gender Gap Index UN Gender Equality Index
(Residualized using Log GDP p/c) (Residualized using Log GDP p/c)

E Ratio Female to Male LFP F Time since Women’s Suffrage

2
2

BOL BOL
Average Gender Difference (Index)
Average Gender Difference (Index)

(Residualized using Log GDP p/c)

(Residualized using Log GDP p/c)

ZWE ZWE
CAN CAN
UKR
USA USA UKR
EST
GBR GBR EST
1
1

DZA CZE AUS SWE UGA DZA UGA SWE CZE AUS
NLD
AUTRUS NLD
AUT RUS
ARG CHN RWA
MWI ARG
RWA
CHNIND
MWI
EGY MARIND NGA LTU
FIN
MAR EGY
NGA LTU
FIN
CHE HTI CHE HTI
IDN POL IDN POL GEO
GEO BWA BWA
AFG HUN COL AFG HUN
SUR PHL COL SRB MDA MDA SUR
SRB PHL
0
0

SAU JPN PER CMR

VNM JPNPER
KEN
CMR
VNM DEU
BGD DEU KEN BGD
JOR CHL ISR JOR ISR CHL ROU
GTM ROU
HRVPRT HRV GTMPRT
BIH
CRI ESP FRA KHM FRA CRIBIHKHM
VEN ESP
TUR LKA NIC VEN BRA KAZ NIC BRA
TUR KAZ
LKA
TZA TZA ZAF
KOR ZAF
GRC GRCKOR
−1
−1

PAK THA GHA GHA PAK THA

ITA ITA
MEX MEX
IRN IRN
IRQ IRQ
ARE
Slope coefficient: 0.2453 (p=0.012) Slope coefficient: 0.2988 (p=0.002)
−2
−2

−2 −1 0 1 2 −2 −1 0 1 2
Ratio Female to Male LFP Time since Women’s Suffrage
(Residualized using Log GDP p/c) (Residualized using Log GDP p/c)

Fig. 2. Analysis of gender differences in preferences in relation of gender differences in preferences and five indices of gender equality
to economic development conditional on gender equality, and vice after residualizing all variables with respect to log GDP per capita. Indices
versa. Each panel depicts a partial regression plot. (A) Relationship of gender equality are (B) the Gender Equality Index, (C) WEF Global
between the aggregate index of gender differences in preferences and log Gender Gap Index, (D) UN Gender Equality Index, (E) ratio of female to
GDP per capita after residualizing both variables with respect to the male labor force participation, and (F) years since women’s suffrage.
Gender Equality Index. (B to F) Relationship between the aggregate index For corresponding regression evidence, see table S4.

Falk et al., Science 362, eaas9899 (2018) 19 October 2018 4 of 6

R ES E A RC H | R E S EA R C H A R T I C LE

across countries. As suggested by the resource the indifference point between receiving a fixed indicated problems in understanding the word-
hypothesis, greater availability of material re- monetary amount on the day of the survey and ing or the quantitative content of the survey
sources removes the human need of subsistence, a larger amount 12 months later, and (ii) the items. Some respondents suggested rewording,
and hence provides the scope for attending to response to the question “How willing are you which was incorporated through minor adjust-
gender-specific preferences. A more egalitarian to give up something that is beneficial for you ments of some survey items.
distribution of material and social resources en- today in order to benefit more from that in the
ables women and men to independently express future?” Positive reciprocity was elicited by (i) Sampling and selection of respondents
gender-specific preferences. an item asking for the value of a thank-you gift We included the GPS as part of the Gallup World
the respondent is willing to give in return for Poll 2012 through the infrastructure of Gallup
Materials and methods help by a stranger and (ii) the response to the (23). Respondents were sampled to achieve na-
See the supplementary materials for further de- item “When someone does me a favor I am tional representativeness of the resident popu-
tails, including a list of the 76 countries included willing to return it.” Negative reciprocity was lation aged 15 and older. Telephone interviews
in the survey. elicited by responses to the items (i) “If I am were conducted in regions where at least 80%
treated very unjustly, I will take revenge at the of the country’s population is covered by tele-
Experimental selection of survey first occasion, even if there is a cost to do so,” (ii) phone or where it is the customary survey meth-
items and construction of “How willing are you to punish someone who odology. Otherwise, face-to-face interviews were
preference measures treats you unfairly, even if there may be costs for conducted.
Survey items included in the GPS data were you?”, and (iii) “How willing are you to punish The selection of households in countries with
selected in an ex ante experimental validation someone who treats others unfairly, even if there telephone interviews used either a random-digit-
procedure at the Laboratory for Experimental may be costs for you?” Altruism was elicited by (i) dialing method or nationally representative lists
Economics of the University of Bonn in winter the quantitative value in response to the ques- of phone numbers. In countries with face-to-face

Downloaded from http://science.sciencemag.org/ on November 6, 2018

2010–2011. In this procedure, 402 subjects par- tion “Imagine the following situation: Today you interviews, primary sampling units were strati-
ticipated in incentivized-choice experiments and unexpectedly received 1000 euros. How much of fied by population size and/or geography. To
responded to a large set of survey items, which this amount would you donate to a good cause?” select sampled households, we used a random-
were either newly developed or taken from exist- and (ii) the response to the question “How will- route procedure. Respondents were selected ran-
ing surveys (25). ing are you to give to good causes without ex- domly by either the latest-birthday or Kish grid
Incentivized-choice experiments were con- pecting anything in return?” Trust was elicited method.
ducted to obtain an incentivized behavioral by the response to the item “I assume that people
measure for each preference. Risk-taking was have only the best intentions.” For each prefer- REFERENCES AND NOTES
measured as the average response to two mul- ence, the final survey measure was given as the 1. M. Bertrand, in Handbook of Labor Economics, O. Ashenfelter,
tiple price lists in which subjects chose between weighted average of the z-scores of the corre- D. Card, Eds. (Elsevier, 2011), vol. 4b, pp. 1543–1590.
a lottery and varying safe options. Patience was sponding survey items. The weights were calcu- 2. R. Croson, U. Gneezy, Gender Differences in Preferences.
J. Econ. Lit. 47, 448–474 (2009). doi: 10.1257/jel.47.2.448
measured as the average response to two multi- lated as the coefficients in OLS regressions of the 3. E. E. Maccoby, C. N. Jacklin, The Psychology of Sex Differences
ple price lists in which subjects chose between incentivized behavioral measures on the respec- (Stanford Univ. Press, 1974).
receiving a payment on the day of the experiment tive survey items. 4. T. DeLeire, H. Levy, Worker Sorting and the Risk of Death on the
or a larger payment 12 months later. Trust was Job. J. Labor Econ. 22, 925–953 (2004). doi: 10.1086/423159
Selection of countries, translation of 5. T. Buser, M. Niederle, H. Oosterbeek, Gender, Competitiveness,
measured as the average amount sent as a first and Career Choices. Q. J. Econ. 129, 1409–1447 (2014).
mover in two investment games. Altruism was survey items, and pretest
doi: 10.1093/qje/qju009
measured as first-mover behavior in a dictator For the GPS, 76 countries were selected with the 6. D. M. Buss, Psychological sex differences: Origins through
game with a charitable organization as recip- goal of providing representative coverage of the sexual selection. Am. Psychol. 50, 164–168 (1995).
doi: 10.1037/0003-066X.50.3.164
ient. Positive reciprocity was measured as the global population. As a key criterion, the selected 7. D. C. Geary, Male, Female: The Evolution of Human Sex
average amount sent back as a second mover in countries covered all development levels and Differences (American Psychological Association, 1998).
two investment games. Negative reciprocity was geographic regions, including 24 in Europe, 22 8. A. H. Eagly, Sex Differences in Social Behavior: A Social-Role
measured as the average amount invested into in Asia, 1 in Oceania, 14 in Africa, and 15 in the Interpretation (Erlbaum, 1987).
9. A. H. Eagly, W. Wood, The origins of sex differences in human
punishment after unilateral defection of the op- Americas. Further, the selection process aimed behavior: Evolved dispositions versus social roles. Am. Psychol.
ponent in a prisoner’s dilemma and the minimum at maximizing variation along country character- 54, 408–423 (1999). doi: 10.1037/0003-066X.54.6.408
acceptable offer in an ultimatum game. istics (such as language and historical, political, 10. U. Gneezy, K. L. Leonard, J. A. List, Gender Differences in
For each preference, we selected those survey and ecological conditions) and favored culturally Competition: Evidence From a Matrilineal and a Patriarchal
Society. Econometrica 77, 1637–1664 (2009). doi: 10.3982/
items for constructing the GPS that exhibited the distinct and non-neighboring countries. ECTA6690
highest predictive power for the corresponding For each country, the selected survey items 11. E. Duflo, Women Empowerment and Economic Development.
incentivized behavioral measure (25). Formally, were translated into the country’s major languages J. Econ. Lit. 50, 1051–1079 (2012). doi: 10.1257/jel.50.4.1051
for each preference the behavioral measure was by at least three translators for each language. 12. M. Doepke, M. Tertilt, Women’s Liberation: What’s in It for
Men? Q. J. Econ. 124, 1541–1591 (2009). doi: 10.1162/
regressed on different combinations of the sur- A first translator suggested, depending on the qjec.2009.124.4.1541
vey items. The combination that maximized the region of the target language, an English, French, 13. R. Fernandez, A. Fogli, C. Olivetti, Mothers and Sons:
adjusted R2 was then selected for the respective or Spanish version of the item. A second trans- Preference Formation and Female Labor Force Dynamics.
preference. lator conducted the translation into the target Q. J. Econ. 119, 1249–1299 (2004). doi: 10.1162/
0033553042476224
Twelve survey questions were selected for the language. A third translator conducted a transla- 14. R. Fernández, A. Fogli, Culture: An Empirical Investigation of
GPS; these comprised a mixture of qualitative tion back to the original language. If a discrep- Beliefs, Work, and Fertility. Am. Econ. J. Macroecon. 1, 146–177
items, measured on an 11-point Likert scale (0 to ancy occurred, the process was iterated until (2009). doi: 10.1257/mac.1.1.146
10), and quantitative items involving economic all translators agreed. Furthermore, monetary 15. R. Inglehart, The Silent Revolution: Changing Values and
Political Styles Among Western Publics (Princeton Univ. Press,
trade-off decisions. Risk-taking was elicited by amounts used in the survey questions were ad- 1977).
(i) an item determining the indifference point justed to correspond to the same share in the me- 16. R. Inglehart, P. Norris, Rising Tide: Gender Equality and Cultural
between a lottery with 50% chance of winning dian income of the target countries. Change Around the World (Cambridge Univ. Press, 2003).
and receiving a fixed certain payment, and (ii) the The survey items were pretested as part of the 17. I. Almås, A. W. Cappelen, K. G. Salvanes, E. Ø. Sørensen,
B. Tungodden, Willingness to Compete: Family Matters.
response to the item “Please tell me, in general, Gallup World Poll 2012 pretest, conducted at the Manage. Sci. 62, 2149–2162 (2016). doi: 10.1287/mnsc.2015.2244
how willing or unwilling you are to take risks.” end of 2011 in 22 countries with a sample size of 18. J. Haushofer, E. Fehr, On the psychology of poverty. Science
Patience was elicited by (i) an item determining 10 to 15 respondents per country. No respondent 344, 862–867 (2014). doi: 10.1126/science.1232491

Falk et al., Science 362, eaas9899 (2018) 19 October 2018 5 of 6

R ES E A RC H | R E S EA R C H A R T I C LE

19. T. Tanaka, C. F. Camerer, Q. Nguyen, Risk and time preferences: 29. M. Yuki, Intergroup Comparison versus Intragroup 39. D. Cesarini, C. T. Dawes, M. Johannesson, P. Lichtenstein,
Experimental and household survey data from Vietnam. Relationships: A Cross-Cultural Examination of Social Identity B. Wallace, Genetic variation in preferences for giving and risk
Am. Econ. Rev. 100, 557–571 (2010). doi: 10.1257/aer.100.1.557 Theory in North American and East Asian Cultural Contexts. taking. Q. J. Econ. 124, 809–842 (2009). doi: 10.1162/
20. V. McLoyd, Socioeconomic disadvantage and child development. Soc. Psychol. Q. 66, 166–183 (2003). doi: 10.2307/1519846 qjec.2009.124.2.809
Am. Psychol. 53, 185–204 (1998). doi: 10.1037/0003-066X.53.2.185 30. S. Guimond, A. Chatard, D. Martinot, R. J. Crisp, S. Redersdorff, 40. N. Zethraeus et al., A randomized trial of the effect of
21. One may be agnostic about the ultimate determinants of Social comparison, self-stereotyping, and gender differences estrogen and testosteron on economic behavior. Proc. Natl.
gender-specific preferences. They may be acquired through in self-construals. J. Pers. Soc. Psychol. 90, 221–242 (2006). Acad. Sci. U.S.A. 106, 6535–6538 (2009). doi: 10.1073/
cross-culturally universal gender roles, or they may be doi: 10.1037/0022-3514.90.2.221 pnas.0812757106
due to biological and evolutionary differences between 31. S. Guimond et al., Culture, Gender, and the Self: Variations and 41. D. Cesarini et al., Heritability of cooperative behavior in the
women and men. Impact of Social Comparison Processes. J. Pers. Soc. Psychol. trust game. Proc. Natl. Acad. Sci. U.S.A. 105, 3721–3726
22. A. Falk, A. Becker, T. Dohmen, B. Enke, D. Huffman, U. Sunde, 92, 1118–1134 (2007). doi: 10.1037/0022-3514.92.6.1118 (2008). doi: 10.1073/pnas.0710069105
Global Evidence on Economic Preferences. Q. J. Econ. 133, 32. S. Guimond, Psychological similarities and differences between 42. B. Wallace, D. Cesarini, P. Lichtenstein, M. Johannesson,
1645–1692 (2018). doi: 10.1093/qje/qjy013 women and men across cultures. Soc. Personal. Psychol. Heriability of ultimatum game responder behavior. Proc. Natl.
23. For further details on the Gallup World Poll, see www.gallup. Compass 2, 494–510 (2008). doi: 10.1111/j.1751- Acad. Sci. U.S.A. 104, 15631–15634 (2007). doi: 10.1073/
com/analytics/213704/world-poll.aspx. 9004.2007.00036.x pnas.0706642104
24. We note that trust is not a preference but a composite trait, 33. A specific concern involves cross-cultural differences in 43. D. C. Geary, Evolution and developmental sex differences. Curr.
including beliefs about others’ behavior, prosocial preferences, gender-specific social comparison (29–32): Survey respondents Dir. Psychol. Sci. 8, 115–120 (1999). doi: 10.1111/1467-
and preferences for risk-taking. Given its importance, however, we in less developed, less gender-egalitarian countries may 8721.00027
decided to include it in our analysis. The results remained be inclined to define themselves relative to people of the same
unchanged when excluding trust from the analysis; see gender (intragroup social comparison). In contrast, those in
AC KNOWLED GME NTS
supplementary materials. more developed and gender-egalitarian countries may be likely
25. A. Falk, A. Becker, T. Dohmen, D. Huffman, U. Sunde, “The to compare themselves to both women and men (intergroup We thank P. Andre, E. Avis, S. Fiedler, K. Jensen, M. Kosfeld,
preference survey module: A validated instrument for social comparison). As a consequence, gender differences L. Shiferaw, and J. Tungodden for helpful comments; L. Henkel,
measuring risk, time, and social preferences” (IZA Discussion estimated through self-assessments may be understated in R. Suchy, and S. Walter for outstanding research assistance;
Paper 9674, 2016). less developed countries. M. Antony for his support in collecting and managing the GPS data;
26. The raw gender differences without controls are qualitatively 34. P. T. Costa, A. Terracciano, R. R. McCrae, Gender differences and M. Entner and S. Sauter for administrative support. Funding:

Downloaded from http://science.sciencemag.org/ on November 6, 2018

very similar (table S1). in personality traits across cultures: Robust and surprising Supported by the European Research Council through ERC
27. This relationship remained robust when controlling for country- findings. J. Pers. Soc. Psychol. 81, 322–331 (2001). #209214 (A.F.). Competing interests: Authors declare no
specific differences such as geographic, demographic, and doi: 10.1037/0022-3514.81.2.322 competing interests. Data availability: The data on preferences
historical factors (table S2). A complementary analysis 35. D. P. Schmitt, A. Realo, M. Voracek, J. Allik, Why Can’t a Man are deposited at www.briq-institute.org/global-preferences/home.
further showed that gender differentiation in preferences was Be More Like a Woman? Sex Differences in Big Five Personality
driven by those components of economic development that Traits Across 55 Cultures. J. Pers. Soc. Psychol. 94, 168–182 SUPPLEMENTARY MATERIALS
disproportionately benefit females relative to males (table S3). (2008). doi: 10.1037/0022-3514.94.1.168
www.sciencemag.org/content/362/6412/eaas9899/suppl/DC1
28. We also investigated, separately for each preference, the 36. S. H. Schwartz, T. Rubel, Sex differences in value priorities:
Materials and Methods
relationship between gender differences and economic devel- Cross-cultural and multimethod studies. J. Pers. Soc. Psychol.
Supplementary Text
opment conditional on gender equality, and vice versa. For 89, 1010–1028 (2005). doi: 10.1037/0022-3514.89.6.1010
Figs. S1 to S9
each preference, gender differences were found to be strongly 37. An example for the biological role in shaping human
Tables S1 to S14
associated with log GDP per capita conditional on the Gender preferences comes from twin studies (39–42).
References (44–52)
Equality Index (fig. S5). Likewise, gender differences were 38. For instance, our results can be accommodated by approaches
found to be highly associated with the Gender Equality Index suggesting interactions of evolutionary or biological 16 January 2018; accepted 20 August 2018
conditional on log GDP per capita (fig. S6). determinants with the social environment (43). 10.1126/science.aas9899

Falk et al., Science 362, eaas9899 (2018) 19 October 2018 6 of 6

Relationship of gender differences in preferences to economic development and gender
equality
Armin Falk and Johannes Hermle

Science 362 (6412), eaas9899.

DOI: 10.1126/science.aas9899

Assessing gender differences

What contributes to gender-associated differences in preferences such as the willingness to take risks, patience,
altruism, positive and negative reciprocity, and trust? Falk and Hermle studied 80,000 individuals in 76 countries who

Downloaded from http://science.sciencemag.org/ on November 6, 2018

participated in a Global Preference Survey and compared the data with country-level variables such as gross domestic
product and indices of gender inequality. They observed that the more that women have equal opportunities, the more
they differ from men in their preferences.
Science, this issue p. eaas9899

ARTICLE TOOLS http://science.sciencemag.org/content/362/6412/eaas9899

SUPPLEMENTARY http://science.sciencemag.org/content/suppl/2018/10/17/362.6412.eaas9899.DC1
MATERIALS

REFERENCES This article cites 34 articles, 5 of which you can access for free
http://science.sciencemag.org/content/362/6412/eaas9899#BIBL

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Use of this article is subject to the Terms of Service

Science (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement of
Science, 1200 New York Avenue NW, Washington, DC 20005. 2017 © The Authors, some rights reserved; exclusive
licensee American Association for the Advancement of Science. No claim to original U.S. Government Works. The title
Science is a registered trademark of AAAS.