Community trust reduces myopic decisions of
Jon M. Jachimowicz
, Salah Chafik
, Sabeth Munrat
, Jaideep C. Prabhu
, and Elke U. Weber
Columbia Business School, Center for Decision Sciences, Columbia University, New York, NY 10025;
Research and Evaluation Division, BRAC, Dhaka 1212,
Judge Business School, University of Cambridge, Cambridge CB2 1AG, United Kingdom; and
Woodrow Wilson School, Andlinger Center for
Energy & Environment, and Department of Psychology, Princeton University, Princeton, NJ 08544
Edited by Carol S. Dweck, Stanford University, Stanford, CA, and approved March 16, 2017 (received for review November 9, 2016)
Why do the poor make shortsighted choices in decisions that
involve delayed payoffs? Foregoing immediate rewards for larger,
later rewards requires that decision makers (i) believe future pay-
offs will occur and (ii) are not forced to take the immediate reward
out of financial need. Low-income individuals may be both less
likely to believe future payoffs will occur and less able to forego
immediate rewards due to higher financial need; they may thus
appear to discount the future more heavily. We propose that trust
in one’s community—which, unlike generalized trust, we find does
not covary with levels of income—can partially offset the effects
of low income on myopic decisions. Specifically, we hypothesize
that low-income individuals with higher community trust make
less myopic intertemporal decisions because they believe their
community will buffer, or cushion, against their financial need.
In archival data and laboratory studies, we find that higher levels
of community trust among low-income individuals lead to less
myopic decisions. We also test our predictions with a 2-y commu-
nity trust intervention in rural Bangladesh involving 121 union
councils (the smallest rural administrative and local government
unit) and find that residents in treated union councils show higher
levels of community trust and make less myopic intertemporal
choices than residents in control union councils. We discuss the
implications of these results for the design of domestic and global
policy interventions to help the poor make decisions that could
Low-income individuals are more likely to make myopic deci-
sions that favor the short-term but neglect long-term out-
comes (1, 2). People living in poverty are more likely to discount
future payoffs compared with wealthier individuals, which can in
part be attributed to the specific environment in which these
decisions are made. From US households (3) to rural Ethiopian
farmers (4), lower wealth predicts higher temporal discount
rates. A myopic orientation, in turn, makes it less likely indi-
viduals escape poverty as they fail to engage in behaviors that
benefit them in the long term, such as investing in education,
health, and finances (1, 5, 6). This creates a vicious cycle: Poverty
leads to short-sighted choices that in turn lead to poverty (7). But
why are the poor more likely to make myopic decisions, and what
interventions can be designed to shift their decisions toward the
Three broad theoretical perspectives address why poor people
appear myopic. An economic perspective views the poor as
people who, like the rest of society, engage in actions that align
with their goals in a rational manner (8, 9). Poor people make
myopic decisions, then, because they lack the opportunities to
alleviate their impoverished situation. They do the best they can,
given their circumstances. A sociological perspective describes
the decisions of the poor as emanating from a “culture of pov-
erty”that often entails misguided goals and motives (10, 11).
Low-income individuals make decisions contrary to their long-
term interests because they value different ends. Finally, a re-
cently proposed psychological perspective suggests that poverty
affects how the poor process information (7). Because poverty-
related concerns consume mental resources, they leave less ca-
pacity for other tasks. This in turn promotes higher discounting
because poor people are not able to adequately plan for the
future (1, 2). Common to all three perspectives is the assumption
that low- and high-income individuals share a similar calculating
logic when trading off intertemporal choices. They differ in the
reasons provided for why this logic gets skewed, proposing a lack
of opportunities, a lack of education, or limited mental band-
width (1, 2, 8, 10–12).
We suggest a related but different possibility, namely that the
poor are engaged in a different kind of mental calculus. To even
consider accepting a delayed payoff requires both a belief that
the delayed payoff will occur (13, 14) and the ability to forego the
immediate payoff (15). Hence, whereas high-income individuals
may ask, “Is a delayed payoff of $100 worth $85 today?”, low-
income individuals may instead ask, “Do I think I will really get
the delayed payoff?”and “Can I afford to forego the immediate
payoff?”Such pessimism or skepticism may have multiple ori-
gins: adverse past experience with delayed payoffs failing to
materialize or the absence of good experiences to draw from (16)
and the tendency for low-income individuals to worry more
about their immediate needs because these needs loom larger
(17). Intertemporal choice thus not only is a question of dis-
counting delayed payoffs for their distance in time, but also
depends on (i) trusting that delayed payoffs will occur and
(ii) trusting that needs are sufficiently met to enable foregoing
the immediate payoff.
Hence, we focus on a different, currently understudied, ele-
ment of intertemporal decisions—trust—and use it to offer an
alternative explanation that helps integrate and reconcile the
More than 1.5 billion people worldwide live in poverty. Even in
the United States, 14% live below the poverty line. Despite
many policies and programs, poverty remains a domestic and
global challenge; the number of US households earning less than
$2/d nearly doubled in the last 15 y. One reason why the poor
remain poor is their tendency to make myopic decisions. With
reduced temporal discounting, low-income individuals could in-
vest more in forward-looking educational, financial, and social
activities that could alleviate their impoverished situation. We
show that increased community trust can decrease temporal
discounting in low-income populations and test this mechanism
in a 2-y field intervention in rural Bangladesh through a low-cost
and scalable method that builds community trust.
Author contributions: J.M.J., S.C., J.C.P., and E.U.W. designed research; J.M.J., S.C., and
S.M. performed research; J.M.J. and S.C. analyze d data; and J.M.J., J.C.P., and E.U.W.
wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
Data deposition: Data and analysis scripts are available online at https://osf.io/smjsa/.
To whom correspondence should be addressed. Email: email@example.com.
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
www.pnas.org/cgi/doi/10.1073/pnas.1617395114 PNAS Early Edition
three approaches above. Specifically, we argue that choosing
delayed outcomes in intertemporal decisions requires trusting
that future payoffs will occur, as well as trusting that immediate
financial needs will be sufficiently met to make considering the
long term possible. In the absence of trust, it might be rational to
favor the short over the long term [as the economic perspective
suggests (8, 12)]. Increasing trust can help change values, goals,
and motives to favor the long over the short term [as the so-
ciological perspective suggests (10, 11)]. Finally, the presence of
trust may help reduce negative affect and stress, thus improving
the quality of long-term decision making [as the psychological
perspective suggests (1, 2)]. In all three cases, however, trust is
the underlying driver of myopic decisions.
We present evidence from four studies using archival, corre-
lational, experimental, and field data to provide support for the
hypothesis that trust drives intertemporal choices. Further, as we
detail below, we suggest that two types of trust matter: (i) gen-
eralized trust, which extends to the social environment more
generally, increases with income, and influences the belief that
long-term payoffs will occur, and (ii) community trust, which
extends to an individual’s community, does not vary with income,
and influences the belief that financial needs will be sufficiently
met. We specifically highlight the role of community trust and
suggest that interventions designed to increase community trust
among low-income individuals can reduce their myopic behavior,
in turn helping them alleviate their impoverished situation.
Finding 1: Generalized Trust Varies with Level of Income
Investing in a long-term payoff implicitly involves trusting that
promised long-term benefits will materialize (13, 14). Studies
conducted with young children show that when they do not trust
their environment, they are less likely to forego immediate
payoffs (e.g., a small quantity of a desired snack) for a delayed,
larger payoff [e.g., a larger quantity of a desired snack (18)].
Indeed, in a situation where the receipt of a delayed option is not
guaranteed, investing in the short term is likely the rational thing
to do (14). Trust can be seen as a mechanism to deal with the
impacts of unpredictability that helps individuals cope with so-
cial uncertainty and complexity (19). This notion is reflected in
the political science literature, which recognizes “generalized
trust”—“a set of moral values [that] create regular expectations
of regular and honest behavior”(ref. 20, p. 53)—as an important
source of individually and socially valuable outcomes, such as
health and happiness (21). Partly for these reasons, generalized
trust plays an important role in economic growth (22, 23).
Evidence suggests that trust is unequally distributed through-
out society. Trust can be thought of as a belief (24) that emerges
from a number of observations or experiences over time (25).
Individuals with higher incomes are more likely to have favorable
experiences in their lives, whereas those with lower incomes are
more likely to experience violations of trust (13, 26). Much of
what poor people experience (e.g., negative income shocks) re-
inforces a lack of trust in their environment (27). The inter-
temporal decisions of low-income individuals may therefore in
part merely be factoring in the perceived uncertainty of long-
term investments paying off (14). To confirm these predictions,
we analyzed data from the World Values Survey (n=220,145), a
nationally representative survey conducted in almost 100 coun-
tries (28). Generalized trust in this survey is assessed through the
question “Generally speaking, would you say that most people
can be trusted or that you need to be very careful in dealing with
people?”Respondents can choose between two possible options:
“Most people can be trusted,”(coded as 0) and “Need to be very
careful”(coded as 1). Although this single-item, dichotomous
measure of generalized trust is problematic (29), studies have
found it to be related to other valid and relevant variables (30,
31). Income in the survey is self-reported on a scale from 1
(lowest group) to 10 (highest group), with respondents asked to
consider “all wages, salaries, pensions, and other incomes”when
responding. We estimate a logistic regression of income as a
predictor of generalized trust and find that the coefficient of
income is significant (β=−0.07, SE =0.002, P<0.001), in-
dicating that high-income individuals have higher levels of gen-
eralized trust. Thus, low-income individuals may be more
doubtful that a long-term payoff will materialize, which can re-
duce the appeal of a larger, later option.
Finding 2: Financial Needs Vary with Levels of Income
In intertemporal choices, low-income individuals also have to
determine whether their current financial situation allows them
to forego the immediate reward. A staggering proportion of
US households—nearly 50%—are unable to come up with
$2,000 over the course of 1 mo if they need to (15). When levels of
savings are low, as is likely the case for low-income individuals,
they may be unable to forego the smaller, sooner payoff because
they require the money to alleviate their immediate needs (32).
To investigate this, we recruited 285 participants from the
United States who were asked to imagine a situation where they
had to choose between receiving $100 today or $150 in 1 y and
probed to list some of the issues they would consider when making
this decision (see Online Pilot Study for additional details). Par-
ticipants additionally responded to a three-item scale that assessed
financial need [e.g., “Given my current financial constraints, I
need to take $100 today rather than wait for the delayed payoff
($150 in one year)”]. Next, we measured participants’levels of
income and their levels of generalized trust through a six-item
scale (33). Finally, we asked participants which of the two op-
tions they would choose: $100 today or $150 in 1 y.
A total of 116 individuals (40.7%) stated that their current fi-
nancial situation constrained their choice. Unsurprisingly, we find
that levels of income are related to financial need (β=−0.035,
SE =0.006, P<0.001), such that lower income is related to higher
financial need. When we introduce both financial need and income
into a linear regression predicting the choice of delayed ($150 in
1 y) over the immediate option ($100 today), only financial need is
a significant predictor (β=−0.12, SE =0.0013, P<0.001), whereas
income is not (β=0.0011, SE =0.0014, P=0.41). Crucially,
generalized trust is not related to financial need (β=0.03, SE =
0.102, P=0.76): Beliefs regarding whether the long-term payoff
will materialize do not influence participants’evaluation of their
financial situation. Similar to data from the World Values Survey
described above, generalized trust is positively related to income
(β=0.008, SE =0.004, P=0.027).
Taken together, findings 1 and 2 suggest that low-income
individuals are both less likely to believe long-term payoffs will
occur and less able to forego the immediate reward due to
higher financial need. Does this, however, mean that low-
income individuals are doomed to being myopic? We turn to
this question next.
Finding 3: Community Trust Can Act as a Buffer for
Prior research emphasizes the important role of the local com-
munity in influencing the experience of everyday life (21).
Communities even shape an individual’s willingness to take fi-
nancial risks. For instance, one study found that Chinese par-
ticipants were less risk averse than Americans, attributing this
difference to cultural differences between the two groups. “In
socially-collectivist cultures like China, family or other in-group
members will step in to help out any group member who en-
counters a large and possibly catastrophic loss”(ref. 34, p. 1208).
In contrast, in individualistic cultures such as the United States,
individuals who make risky decisions are usually expected to face
the consequences of their decisions. The social structure that
reflects collectivistic societies therefore acts as a “cushion”
against possible losses from risky decisions, allowing individuals
in collectivistic societies to be less risk averse (34, 35). Such
differences exist not only between, but also within, nations (36);
one study suggests that nearly 80% of total cultural variation
exists within, rather than between nations (13).
Supporting evidence for the important role of the commu-
nity also originates from research conducted on the “buffering
www.pnas.org/cgi/doi/10.1073/pnas.1617395114 Jachimowicz et al.
hypothesis”that suggests that strong ties to close others have
beneficial effects on individuals’well-being (37, 38). The per-
ceived availability of support allows individuals to appraise
stressful situations as less aversive, which makes it less likely such
events will negatively influence them (39). Because greater fi-
nancial need is often experienced as stressful (32), stronger
support from the local community may also reduce the aversive
impact of financial need.
Importantly, the experiences that give rise to community trust
are based on an individual’s interactions with his or her imme-
diate surroundings and not with the general environment as a
whole, as is the case for generalized trust. Thus, trust in one’s
local community to cushion against potential losses, or “buffer”
against the stress of lower income, may be distributed more
evenly across the income spectrum than generalized trust. To
further investigate this, we again turn to the World Values
Survey where individuals also respond to the question: “I’d like
to ask you how much you trust people from your neighborhood.
Could you tell me whether you trust people from this group?”
Respondents have four possible options: (i)“Trust completely,”
(ii)“Trust somewhat,”(iii )“Do not trust very much,”or (iv)“Do
not trust at all.”We conducted an ordinal logistic regression of
community trust against levels of income and find that levels of
income do not predict levels of community trust (P=0.15; World
In addition, in the pilot study for finding 2 above (with 285 US
participants), we also measured levels of community trust using a
13-item measure (adapted from ref. 40). Example items included
“I do a lot of good things in my neighborhood”and “There are
advantages to living in my neighborhood.”We regress commu-
nity trust onto our measure of financial need and find a signifi-
cant negative relationship (β=−0.58, SE =0.136, P<0.01), such
that individuals with higher community trust reported lower fi-
nancial need. This relationship also holds when adding income as
an additional predictor (β=−0.40, SE =0.126, P<0.01). Hence,
a higher level of community trust influences the choice of
delayed payoffs by reducing low-income individuals’perceived
constraints. The same level of actual financial need, based on an
individual’s current financial situation, may be experienced dif-
ferently with varying levels of community trust. When individuals
experience lower levels of community trust, actual financial
needs are unlikely to be mitigated by the local community.
However, when individuals experience higher levels of commu-
nity trust, actual financial needs are alleviated by the buffering
and cushioning the local community provides. In turn, when their
needs are not felt as acutely, low-income individuals with higher
community trust are better able to consider the long-term option.
Accordingly, we argue that it is important to consider the ef-
fects of generalized trust as distinct from community trust on the
long-term decisions of the poor. Although a higher level of both
generalized and community trust can theoretically support the
choice of a delayed payoff, we propose there are at least three
reasons why a focus on community trust is a more viable basis for
an intervention to reduce myopic intertemporal choices among
low-income individuals. First, because community trust does not
covary with income whereas generalized trust does, low-income
individuals may already have higher base rates of community
trust, and this may make an intervention simpler and more ef-
fective. Second, personal beliefs are often resistant to change
(41). Whereas interventions to influence personal beliefs exist
(42), they require repeated, in-depth exposure and experiences
that serve to reinforce the intended belief change. Generalized
trust is a more entrenched belief and less amenable to change
than community trust; the latter, based on interactions with one’s
immediate surroundings, has more touchpoints for possible in-
terventions. Moreover, interventions that focus on community
trust require fewer major changes to governmental institutions
compared with treatments that aim to change generalized trust.
Creating an intervention to change generalized trust requires
more time, intense exposure, and systemic change than an in-
tervention to change community trust.
Third, whereas generalized trust influences intertemporal
choices by signaling to individuals how likely it is the long-term
payoff will occur, community trust influences the level of finan-
cial need individuals experience and thus their ability to consider
foregoing the immediate payoff. There may be instances when a
lack of generalized trust is warranted, i.e., where the delayed
payoff—should individuals choose it—does not occur. Thus, an
intervention that increases generalized trust may backfire when
low-income individuals choose the larger, later option and it
does not materialize. Instead, a focus on community trust is less
likely to backfire because its higher levels ameliorate the finan-
cial constraints low-income individuals face. We now turn to
study 1, which seeks to establish the role of community trust in
influencing temporal discounting by low-income individuals.
Study 1: Community Trust and Temporal Discounting by
This study was an online experiment with 647 participants from
the United States (see Supporting Information for additional
details). We first presented respondents with the same 13-item
scale of community trust as above (40). We then assessed their
temporal discount factor (the multiplier that equates $100 in
1y’s time with the amount that an individual is willing to take
instead, if received today) using Dynamic Experiments for Esti-
mating Preferences (DEEP) (43), an adaptive testing platform
where participants repeatedly choose between a smaller payoff
that is received closer to the present (smaller/sooner) and a
larger one that is received farther into the future (larger/later).
Although decisions are hypothetical, temporal discount factors
predict real-world intertemporal decisions, such as mortgage
choices (44), and their consequences, such as credit scores (45).
Indeed, decisions in other delay-discounting tasks are predictive
of a wide range of long-term outcomes, such as health, educa-
tion, and retirement savings (46, 47). Finally, respondents
reported their current levels of income, as well as their gender,
age, and education.
Replicating previous studies (1, 2, 7), we find that discount
factors vary with levels of income (β=0.0034, SE =0.0015, P=
0.021), such that individuals with higher levels of income dis-
counted the future less than those with lower levels of income.
To illustrate this, we categorized participants with household
incomes below $40,000 as low income and those with incomes
above $40,000 as high income and found that low-income par-
ticipants discounted the future more (M=0.131, SE =0.006)
than high-income participants (M=0.159, SE =0.006), with
lower discount factors indicating greater discounting. This cutoff
point represents the median in our sample. Similar cutoff points
are often used in prior research (48, 49). Cutoff points higher or
lower than $40,000 do not significantly change our results.
Next, we regressed the discount factor on continuous income
and community trust as well as the interaction between the two
predictor variables. In addition to the main effect of income
already mentioned, we find a main effect of community trust
(β=0.0015, SE =0.0007, P=0.025), such that individuals with
higher levels of community trust discount the future less. Both
main effects are qualified by a significant interaction between
community trust and levels of income on temporal discounting
(β=−0.0052, SE =0.0023, P=0.026). These effects also hold
when we control for demographic variables such as age, gender,
and education (Table S1). To better understand the interaction
between community trust and income, we next conducted simple
slopes analyses (50) and found that lower levels of income were
related to higher discounting of the future only when levels of
community trust were low (t(643) =2.86, P=0.0044) but not
when levels of community trust were high (t(643) =−0.032, P=
0.748). Hence, only individuals with low incomes and low levels
of community trust differ significantly from all other groups (Fig.
S1). We now turn to study 2, which seeks to examine the impact
of community trust on the temporal discounting of low-income
individuals in a richer real-world context.
Jachimowicz et al. PNAS Early Edition
Study 2: Community Trust and Payday Loans
In study 2 we investigate whether taking out a payday loan—a
typical form of myopic behavior displayed by low-income indi-
viduals—varies with levels of community trust (see Supporting
Information for additional details). To do so, we combine state-
level data from the Survey of Household Economics and De-
cision Making (SHED) with an additional survey that measured
community trust, which we conducted among 5,721 US partici-
pants in 50 states. We recruited US participants through a
stratified sampling method, such that ∼100 participants respon-
ded per state. Participants responded to questions assessing their
levels of community trust, using the same scale as in study 1.
Based on these responses, we created state averages. We also
obtained state-level data of additional control variables, such as
income, unemployment, and age. Through SHED, we accessed
state-level data on payday loan use and matched both datasets at
the state level.
An ordinary least-squares regression with state-level payday
loan use as the dependent variable and state-level community
trust as the independent variable finds that community trust
predicts payday loan use (β=−0.15, SE =0.041, P<0.001). This
effect also holds when we control for other variables such as age,
income, and unemployment. Crucially, this effect also holds
when controlling for levels of savings (β=−0.11, SE =0.033, P=
0.001), a proxy for levels of actual financial need. This provides
further support that higher levels of community trust reduce
perceived financial need, even when levels of actual financial
need vary. Although studies 1 and 2 suggest that community trust
plays a role in buffering or cushioning low-income individuals
against myopic discounting, this evidence is correlational. We
now turn to a study that attempts to establish causal evidence for
the proposed relationship.
Study 3: Exploring the Causal Link between Community
Trust and Temporal Discounting by Low-Income Individuals
in the Laboratory
We recruited 120 participants online and assigned them to one
of four possible conditions in a 2 ×2 design. Specifically, the
design involved manipulating levels of felt income (low/high) and
levels of felt community trust (low/high). Imagining more severe
financial implications has been shown to evoke feelings of having
lower income (2). To induce low vs. high levels of felt income, we
used previously developed and validated scenarios (2). Partici-
pants in the high felt-income condition were asked to imagine
scenarios with relatively minor financial implications, whereas
those in the low felt-income condition were asked to imagine
scenarios with more severe financial implications. We manipu-
lated levels of community trust by increasing the salience of this
construct in the minds of respondents (51). We gave participants
a definition of community trust (“the extent to which you trust
your community”). We then asked them to list either 2 (low) or
10 (high) examples from their own experience where community
trust was justified. In contrast to studies that use a similar design
to manipulate difficulty of retrieval (52), participants in this
study had to produce the full number of examples requested.
Subjects did not experience difficulties in providing examples.
Next we assessed temporal discounting, using DEEP (43). We
also collected data on several demographic variables.
Consistent with what we would expect if our manipulation of
felt income was successful, we found that participants in the low
felt-income condition were more myopic (M=0.13, SE =0.015)
than participants in the high felt-income condition (M=0.178,
SE =0.017, P=0.044). We examined whether community trust
serves as a buffer or cushion for individuals with lower levels of
felt income by testing for an interaction effect between levels of
community trust and felt income on the temporal discount fac-
tor. An ANOVA with felt income and manipulated community
trust as the independent variables and the discount factor as the
dependent variable shows a marginally significant interaction
=2.98, P<0.10). To further investigate which condition
is driving this effect, we conducted pairwise comparisons. These
revealed that three conditions differ significantly from a fourth.
Participants in the low felt-income, low community-trust condi-
tion were more myopic (M=0.103, SE =0.019) than individuals
in the low felt-income, high community-trust (M=0.178, SE =
0.024; P=0.04), high felt-income, low community-trust (M=
0.176, SE =0.018; P=0.032), and high felt-income, high
community-trust (M=0.179, SE =0.03; P=0.045) conditions.
These results hold when controlling for additional control vari-
ables (e.g., age, gender, education, and actual income).
Study 3 provides laboratory-based causal evidence in support
of our hypothesis that low-income individuals with higher levels
of community trust discount the future less heavily than low-
income individuals with lower levels of community trust. Al-
though our previous studies show that community trust does not
vary by income and that perceptions of such trust can be ma-
nipulated in a laboratory setting, we now turn to showing that
community trust can be built in a real-world context and test if
doing so reduces myopic intertemporal decisions.
Study 4: Exploring the Causal Link Between Community
Trust and Temporal Discounting by Low-Income Individuals
in the Field
In this study, we sought to replicate our findings in a field setting
featuring a different cultural context and involving ultrapoor
individuals (see Supporting Information for additional details).
We collaborated with BRAC, an international development or-
ganization based in Bangladesh, and The Hunger Project (THP),
a global nonprofit organization with headquarters in New York.
In February 2014, BRAC and THP launched a 2-y intervention
designed to increase community trust in 121 union councils (the
smallest rural administrative and local government units) in four
districts of Bangladesh (Kishoreganj, Habiganj, Sunamgonj, and
Bagerhat). Sixty-one union councils received the intervention
whereas 60 union councils were in the control condition (see Table
S2 for demographic information and Supporting Information for
The intervention had two components. First, volunteers from
the community were trained to act as intermediaries between the
community and the local government. This required the volun-
teers to interact with other members of their community, provide
input into local governance, and help residents access public
services from the local government. Second, a platform was
created for inclusive community-driven governance to change
the way community-level decisions were made. This involved
representatives from the community working with the local
government to make community-level decisions, for example in
the distribution of social benefits, the allocation of funds and
resources for development projects, and the selection of people
to use in publicly funded projects. At the end of the 2-y in-
tervention, we surveyed individuals (n=1,447) in all 121 union
councils on their levels of community trust as well as assessing
their temporal discounting. We measured temporal discounting
using a pen-and-paper titration measure (53).
We first tested whether our intervention increased levels of
community trust in treatment union councils. Our intervention
was successful: We find a significant difference in levels of
community trust between treatment and control union councils
(β=−0.14, SE =0.026, P<0.001), such that levels of community
trust (ranging from 1 to 5) are higher in treatment (M=3.45,
SE =0.0015) than control union councils (M=3.31, SE =
0.0013). There were no significant differences between treatment
and control union councils for generalized trust. We next spec-
ified a hierarchical linear model, which nests union councils
within condition and clusters SEs at the union council level. This
allows us to account for differences between union councils and
provides a more accurate analysis of the treatment effect. Our
dependent variable is individuals’temporal discount factor. As
Table S3 shows, participants in treatment union councils were
significantly more likely to discount the future less heavily (β=
0.081, SE =0.034, P=0.018). In concordance with our prior
studies, measured generalized trust, as shown in model 2 in
www.pnas.org/cgi/doi/10.1073/pnas.1617395114 Jachimowicz et al.
Table S3, is an additional significant predictor (β=0.54, SE =
0.13, P<0.01), such that those individuals with higher levels of
generalized trust are more likely to discount the future less. The
addition of further control variables does not significantly in-
fluence individuals’tendency to discount the future (model 3 in
Table S3). This effect also holds when controlling for levels of
income, a proxy for levels of actual financial need, providing
further evidence that higher levels of community trust reduce
perceived financial need even when levels of actual financial
need vary. To further establish the role of community trust in
reducing perceived financial need, we also conducted 42 quali-
tative interviews and 8 focus group discussions in 14 union
councils, 7 treatment and 7 control (see Supporting Information
for further information).
In sum, this field study shows that an intervention designed to
increase levels of community trust successfully does so and, in
the process, affects temporal discounting, such that individuals in
treatment union councils are less myopic in their intertemporal
decisions than individuals in control union councils.
Low-income individuals are more likely to make myopic deci-
sions. This can, in turn, make it more difficult for them to alle-
viate their impoverished condition. At least three broad
perspectives have addressed why low-income individuals are
more likely to discount the future more heavily. An economic
perspective views individuals living in poverty as people who, like
the rest of society, engage in actions that align with their goals in
a rational manner (8, 9). A sociological perspective describes the
decisions of the poor as emanating from a culture of poverty that
often entails misguided goals and motives (10, 11). Finally, a
psychological perspective suggests that poverty itself affects in-
dividuals’information processing (7). These perspectives share
the assumption that low- and high-income individuals use a
similar logic in their trade-off calculation.
In this paper, we focus on a different, understudied, element
of intertemporal decisions—trust. We show that low-income
individuals are more likely to make myopic decisions because
(i) they have lower levels of generalized trust, thus reducing their
belief that the delayed payoff will occur, and (ii) they have higher
levels of financial need, thus constraining their ability to forego
the immediate payoff. Because community trust reduces the felt
impact of actual financial need, low-income individuals with
higher levels of community trust make less myopic intertemporal
decisions. Indeed, community trust reduces myopic inter-
temporal choices even when controlling for actual financial need
as in studies 2 and 4, providing further support that higher levels
of community increase levels of perceived financial need. By
increasing levels of community trust, the myopic behavior of low-
income individuals can be reduced, potentially helping them
improve their financial well-being. Generalized trust, in our
studies as well as in previous work, also affects people’s delay
discounting but may be more difficult to change. It is worth
noting that our community trust intervention in study 4 did not
impact levels of generalized trust.
This paper makes three primary contributions. First, we
highlight that aside from the differential impact of time delay,
intertemporal choice may also be influenced by beliefs about
whether long-term payoffs will occur and the ability to forego
immediate payoffs. Because low-income individuals are less
likely to generally trust their environment, myopic decisions may
reflect not just greater impatience, but also reduced belief that
long-term payoffs will occur. In addition, because low-income
individuals are more likely to experience greater financial need,
myopic decisions may also reflect an inability to consider long-
term options. This perspective allows us to integrate previous
approaches that have attempted to explain why low-income in-
dividuals are more likely to discount the future more heavily and
provides a single consistent explanation capable of reconciling
differences between approaches. Specifically, in the absence of
trust, it might be rational to favor the short over the long term (as
the economic perspective suggests). Further, the presence of trust
can help reduce negative affect and stress, in turn improving the
quality of long-term decision making (as the psychological
perspective suggests). Increasing trust can help change values,
goals, and motives to favor the long term over the short term
(as the sociological perspective suggests). Those low-income
individuals who trust their community may be more willing to
choose delayed payoffs because they are able to rely on their
community to alleviate their financial needs, which in turn al-
lows them to consider foregoing immediate payoffs. In all cases,
trust is an underlying driver of the change in myopic behavior.
Second, we distinguish between generalized trust, which we
and others show to vary with income, and community trust,
which we show does not. Because community trust deals only
with an individual’s immediate social environment, and not with
the general environment as a whole, interventions need only
focus on an individual’s direct social environment, rather than
the general environment as a whole. Generalized trust reflects a
more enduring mindset, whereas beliefs about one’s community
are drawn from people’s transactions and interactions with their
immediate surroundings, which are more amenable to targeted
interventions. Third, our theoretical model generates a unique
intervention strategy that we tested in the context of rural Ban-
gladesh. Specifically, an intervention designed to increase levels
of community trust was effective in shifting temporal preferences
toward the long term. Such an approach has benefits over in-
terventions based on prior perspectives on the myopic behavior
of low-income individuals that have produced mixed results, for
example through microfinance (54) or financial literacy pro-
grams (55). In contrast, because higher community trust reduces
perceived financial need, this paper highlights a relatively low-
cost, empowering, and scalable intervention.
Whereas each of our studies has its individual limitations, we
deliberately adopted a multiple-study strategy that varies meth-
ods, types of data, and contexts to ensure that the strengths of
each study would compensate for the weaknesses of the others
and that, taken together, they would generate broad support for
our theoretical model. Thus, in our laboratory and field studies,
we focus on temporal discounting but do not examine whether
changes in temporal discounting lead to changes in downstream
behavior. However, study 2 shows that our model holds when
predicting real-world payday loan use. And whereas this archival
study did not use individual-level data, we attempted to provide
that level of rigor in our controlled experimental laboratory
studies. Finally, whereas our laboratory studies lack external
validity, we aim to provide this through our 2-y field study that
manipulates levels of community trust in rural Bangladesh. Due
to field constraints, we were unable to collect data from the same
individuals before and after the intervention in Bangladesh.
Doing so would have allowed for a more powerful research de-
sign including a difference-in-difference comparison (56). We
also did not incentivize our intertemporal choice tasks. Although
it is preferable to use incentivized tasks, hypothetical choice tasks
are widely used and predictive of real-world outcomes (45).
Future research should incorporate a repeated-measures design
that incentivizes intertemporal choices before and after in-
tervention and tracks the impacts of the intervention for im-
portant real-world outcomes, such as levels of income over time.
Poverty is one of the world’s most vexing problems. Although
great progress has been made in alleviating poverty, there is still
a long way to go, both domestically and globally. For example,
in the United States, the number of households with less than
$2/d per person has nearly doubled in the last 15 y (57). Progress
is often impeded because low-income individuals tend to dis-
count the future more than is advised. To tackle this challenge,
our theory and results suggests policy should move beyond
a sole focus on the low-income individual and instead pro-
vide additional emphasis on the low-income community. Policy
makers could implement changes that give individuals in low-
income communities more opportunities to develop commu-
nity trust. This can be achieved, for example, by increasing the
Jachimowicz et al. PNAS Early Edition
opportunities for interaction or giving community members
more say over decision making at the local level. The poor may
lack in material wealth relative to the rich, but they possess
social wealth in the shape of their communities upon which
they can draw. Building and boosting community trust can help
decrease myopic decision making and, in turn, contribute to
reducing the incidence of poverty domestically and worldwide.
All experiments we report here were approved by the Co-
lumbia University Institutional Review Board and all partici-
pants provided informed consent.
ACKNOWLEDGMENTS. We thank Geun Hae Ahn for research assistance;
Tuhin Alam, Andrew Jenkins, Maria May, and the rest of the team at BRAC
and The Hunger Project (THP) for field support; and Christina Boyce-Jacino,
Andrea Dittmann, Lilly Kofler, Stephan Meier, Rachel Meng, Anuj Shah, the
White House Social and Behavioral Science Team, and participants at the
Society for Personality and Social Psychology Conference and the Judgment
and Decision-Making Conference for helpful comments on earlier drafts.
This research was made possible in part by a Cambridge Judge Business
School small grant, the research facilities provided by the Center for Decision
Sciences at Columbia University, and the support of the German National
1. Mani A, Mullainathan S, Shafir E, Zhao J (2013) Poverty impedes cognitive function.
2. Shah AK, Mullainathan S, Shafir E (2012) Some consequences of having too little.
3. Lawrance EC (1991) Poverty and the rate of time preference: Evidence from panel
data. J Polit Econ 119:54–77.
4. Yesuf M, Bluffstone R (2008) Wealth and time preference in rural Ethiopia. EfD
Discuss Pap 23:1–19.
5. Lopez RB, Hofmann W, Wagner DD, Kelley WM, Heatherton TF (2014) Neural pre-
dictors of giving in to temptation in daily life. Psychol Sci 25:1337–1344.
6. Casey BJ, et al. (2011) Behavioral and neural correlates of delay of gratification
40 years later. Proc Natl Acad Sci USA 108:14998–15003.
7. Haushofer J, Fehr E (2014) On the psychology of poverty. Science 344:862–867.
8. Sen AK (1985) Commodities and Capabilities (Oxford Univ Press, Oxford).
9. Sen AK, Nussbaum M (1993) The Quality of Life (Oxford Univ Press, Oxford).
10. Lewis O (1959) Five Families: Mexican Case Studies in the Culture of Poverty (Basic
Books, New York).
11. Lewis O (1966) La Vida: A Puerto Rican Family in the Culture of Poverty - San Juan and
New York (Wiley, New York).
12. Sen AK (1987) On Ethics and Economics (Blackwell, Oxford).
13. Dohmen T, Enke B, Falk A, Huffman D, Sunde U (2016) Patience and the wealth of
nations. Working papers 2016-012 (Human Capital and Economic Opportunity Working
Group, Chicago). Available at econpapers.repec.org/paper/hkawpaper/2016-012.htm.
14. Gabaix X, Laibson D (2017) Myopia and discounting. Working paper no. 23254 (National
Bureau of Economic Research, New York). Available at www.nber.org/papers/w23254.
15. Lusardi A, Schneider D, Tufano P (2011) Financially fragile households: Evidence and
implications. Brookings Pap Econ Act 1:83–134.
16. Uslaner EM (2008) Where you stand depends upon where your grandparents sat: The
inheritability of generalized trust. Public Opin Q 72:725–740.
17. Shah AK, Shafir E, Mullainathan S (2015) Scarcity frames value. Psychol Sci 26:402–412.
18. Michaelson L, de la Vega A, Chatham CH, Munakata Y (2013) Delaying gratification
depends on social trust. Front Psychol 4:355.
19. Earle TC, Cvetkovich G (1995) Social Trust: Toward a Cosmopolitan Society (Praeger,
20. Fukuyama F (1995) Trust: The Social Virtues and Creation of Prosperity (Hamish
21. Putnam R (2000) Bowling Alone. The Collapse and Revival of American Community
(Simon and Schuster, New York).
22. Knack S, Keefer P (1997) Does social capital have an economic payoff? A cross-country
investigation. Q J Econ 112:1251–1288.
23. Algan Y, Cahuc P (2010) Inherited trust and growth. Am Econ Rev 100:2060–2092.
24. Falk A, et al. (2015) The nature and predictive power of preferences: Global evidence.
Discussion paper no. 9504 (Institute for the Study of Labor, Bonn, Germany). Available
25. Evans EM (2001) Cognitive and contextual factors in the emergence of diverse belief
systems: Creation versus evolution. Cognit Psychol 42:217–266.
26. Rothstein B, Uslaner EM (2005) All for all: Equality, corruption, and social trust. World
27. Brandt MJ, Wetherell G, Henry PJ (2015) Changes in income predict change in social
trust: A longitudinal analysis. Polit Psychol 36:761–768.
28. World Values Survey Association (2014) WORLD VALUES SURVEY Wave 6 2010–2014
OFFICIAL AGGREGATE v.20150418 (Vienna).
29. Miller AS, Mitamura T (2003) Are surveys on trust trustworthy? Soc Psychol Q 66:
30. Rothstein B, Stolle D (2008) The state and social capital: An institutional theory of
generalized trust. J Comp Polit 40:441–459.
31. Bjørnskov C (2007) Determinants of generalized trust: A cross-country comparison.
Public Choice 130:1–21.
32. Ruberton PM, Gladstone J, Lyubomirsky S (2016) How your bank balance buys hap-
piness: The importance of “cash on hand”to life satisfaction. Emotion 16:575–580.
33. Yamagishi T, Yamagishi M (1994) Trust and commitment in the United States and
Japan. Motiv Emot 18:9–66.
34. Weber EU, Hsee CK (1998) Cross-cultural differences in risk perception but cross-
cultural similarities in attitudes towards risk. Manage Sci 44:1205–1212.
35. Hsee C, Weber E (1999) Cross-national differences in risk preference and lay predic-
tions. J Behav Decis Making 12:165–179.
36. Harrington JR, Gelfand MJ (2014) Tightness-looseness across the 50 United States.
Proc Natl Acad Sci USA 111:7990–7995.
37. Cohen S, Wills TA (1985) Stress, social support, and the buffering hypothesis. Psychol
38. Cobb S (1976) Presidential Address-1976. Social support as a moderator of life stress.
Psychosom Med 38:300–314.
39. Baumeister RF, Leary MR (1995) The need to belong: Desire for interpersonal at-
tachments as a fundamental human motivation. Psychol Bull 117:497–529.
40. Ramos D, Victor T, Seidl-de-Moura ML, Daly M (2013) Future discounting by slum-
dwelling youth versus university students in Rio de Janeiro. J Res Adolesc 23:95–102.
41. Howe PD, Mildenberger M, Marlon JR, Leiserowitz A (2015) Geographic variation in
opinions on climate change at state and local scales in the USA. Nat Clim Chang 5:
42. Ball-Rokeach SJ, Rokeach M, Grube JW (1984) The Great American Values Test:
Influencing Behavior and Belief Through Television (Free Press, New York).
43. Toubia O, Johnson E, Evgeniou T, Delquié P (2013) Dynamic experiments for esti-
mating preferences: An adaptive method of eliciting time and risk parameters.
Manage Sci 59:613–640.
44. Johnson EJ, Atlas SA, Payne JW (2016) Time preferences and mortgage choice. J Mark
Res, in press.
45. Li Y, et al. (2015) Sound credit scores and financial decisions despite cognitive aging.
Proc Natl Acad Sci USA 112:65–69.
46. Weber EU, et al. (2007) Asymmetric discounting in intertemporal choice: A query-
theory account. Psychol Sci 18:516–523.
47. Berns GS, Laibson D, Loewenstein G (2007) Intertemporal choice–toward an in-
tegrative framework. Trends Cogn Sci 11:482–488.
48. Carvalho LS, Meier S, Wang SW (2016) Poverty and economic decision-making: Evi-
dence from changes in financial resources at payday. Am Econ Rev 106:260–284.
49. Kearney M, Harris B, Jáconne E, Parker L (2013) A Dozen Facts about America’s
Struggling Lower-Middle Class (The Brookings Institution, Washington, DC).
50. Preacher KJ, Curran PJ, Bauer DJ (2006) Computational tools for probing interactions
in multiple linear regression, multilevel modeling, and latent curve analysis. J Educ
Behav Stat 31:437–448.
51. Strack F, Martin LL, Schwarz N (1988) Priming and communication: Social determi-
nants of information use in judgments of life satisfaction. Eur J Soc Psychol 18:
52. Schwarz N, Bless H, Strack F, Klumpp G (1991) Ease of retrieval as information: An-
other look at the availability heuristic. J Pers Soc Psychol 61:195–202.
53. Green L, Fry AF, Myerson J (1994) Discounting of delayed rewards: A life-span com-
parison. Psychol Sci 5:33–36.
54. Banerjee A, Karlan D, Zinman J (2015) Six randomized evaluations of microcredit:
Introduction and further steps. Am Econ J Appl Econ 7:1–21.
55. Fernandes D, Lynch JG, Netemeyer RG (2014) Financial literacy, financial education
and downstream financial behaviors. Manage Sci 60(8):1–23.
56. Donald SG, Lang K (2007) Inference with difference-in-differences and other panel
data. Rev Econ Stat 89:221–233.
57. Edin K, Shaefer HL (2015) $2.00 a Day (Houghton Mifflin, New York).
58. Brehm J, Rahn W (1997) Individual-level evidence for the causes and consequences of
social capital. Am J Pol Sci 41:999–1023.
59. Laibson D (1997) Golden eggs and hyperbolic discounting. Q J Econ 112:443–478.
60. Federal Deposit Insurance Corporation (2015) FDIC National Survey of Unbanked and
Underbanked Households (Federal Deposit Insurance Corporation, Washington, DC).
61. Consumer Federation of America (2016) Legal Status of Payday Loans by State
(Consumer Federation of America, Washington, DC).
62. Greene WH (1993) Econometric Analysis (Macmillan, New York), 2nd Ed.
63. Li Y, Baldassi M, Johnson EJ, Weber EU (2013) Complementary cognitive capabilities,
economic decision making, and aging. Psychol Aging 28:595–613.
64. Ahmed T, Munrat S, Naher K (2017) An Evaluation Study on BRAC-THP Strengthening
Local Governance Project (BRAC, Dhaka, Bangladesh).
www.pnas.org/cgi/doi/10.1073/pnas.1617395114 Jachimowicz et al.
Jachimowicz et al. 10.1073/pnas.1617395114
World Values Survey
The World Values Survey (WVS) is a collection of surveys
concerning human beliefs and values that is conducted in nearly
100 countries, together covering 90% of the global population.
Levels of generalized trust are assessed by “Generally speaking,
would you say that most people can be trusted or that you need
to be very careful in dealing with people?”There are only two
possible responses: (i)“Most people can be trusted”or (ii)
“Need to be very careful.”The WVS assesses levels of commu-
nity trust with the following question: “I’d like to ask you how
much you trust people from your neighborhood. Could you tell
me whether you trust people from this group?”There are four
possible response options: (i)“Trust completely,”(ii)“Trust
somewhat,”(iii)“Do not trust very much,”or (iv)“Do not trust
at all.”Income is measured via a scale from 1 (lowest income
group) to 10 (highest income group) “in what group your
household is.”Survey participants are asked to consider “all
wages, salaries, pensions and other incomes that come in”when
specifying the appropriate number. Although the correlation
between generalized and community trust is significant (r=
0.328, P≤0.001), the divergent relationship to income provides
support for the view that the two sources of trust can be thought
of as distinct.
Online Pilot Study
We recruited 285 participants (127 females, mean age =35.26 y,
SD =11.36) through Amazon.com’s Mechanical Turk (MTurk)
website. We first measured individuals’levels of income and
their levels of generalized trust, using a 6-item scale (adapted
from ref. 33) (e.g., “Most people are trustworthy”and “Iam
trustful”;α=0.90). We used this 6-item measure of generalized
trust with seven scale points to address potential validity con-
cerns of the 1-item, dichotomous scale used in the WVS (33).
This 6-item scale has been found to have sufficient validity and
reliability and has been widely used in prior research (58). Par-
ticipants were also asked to respond to a measure of community
trust, using a 13-item scale (adapted from ref. 40) [e.g., “I would
like my child(ren) to be raised in the neighborhood I currently
live”and “I do a lot of good things in my neighborhood”;α=
0.82]. Then we asked participants to imagine they were given a
choice between receiving $100 today or $150 in 1 y and assessed
their level of financial need, using a 3-item scale [“Given my
current financial constraints, I need to take $100 today rather
than wait for the delayed payoff ($150 in one year)”;“If I could, I
would choose the delayed payoff ($150 in one year), but my
current financial situation does not allow me to”; and “I want to
choose the delayed payoff ($150 in one year), but I need to
choose the close one ($100 today)”;α=0.84]. Finally, partici-
pants were asked to choose which one of the two options they
would prefer, $100 today or $150 in 1 y.
Study 1: Correlational Evidence from the United States
We recruited 647 participants (319 females, mean age =35.55 y,
SD =10.63) through Amazon.com’s MTurk website. Following
informed-consent procedures, participants responded on a five-
point scale ranging from “not at all”to “completely”to the same
13-item measure of community trust as above (α=0.87).
We next assessed temporal discounting factors using DEEP
(43), an adaptive testing framework where participants choose
between a payoff that is smaller but is received closer to the
present (smaller sooner) and one that is larger but is received
farther into the future (larger later). The structure of each de-
cision depends on the participant’s previous response, allowing
for a precise, robust, and fast time preference estimation. Markov
chain Monte Carlo and hierarchical Bayesian statistics are used to
determine more accurate values of the time function—specifically
beta (present bias) and delta (discounting rate). As part of a final
section on demographic information, participants indicated their
current levels of income, as well as their age, education, and sex.
Similar to previous research, the temporal discounting factor
was computed by first calculating the annual discounting rate
(ADR) (exp(δ* 365)) and then calculating the discounting factor
as 1/1(1 +ADR). Let δdays be the discounting factor in day units.
The discount factor in a βδ model hyperbolic discounting framework
DFdays =βδDdays .
where Ddays is time periods in days:
Considering one time-period delay (whatever unit), then
For hyperbolic discounting,
for interest rate r, and
for exponential discounting.
Therefore, rhyperbolic =1−δ
δand rexponential =−logðδÞ.
Study 2: Archival Payday Loan Data
We recruited 5,721 participants (51% female; age M=38.01 y;
SD =2.16) on MTurk across all 50 US states. We committed to a
sample size requirement in advance of launching our study. We
aimed to have equal representation of all states in our dataset.
Our target sample size was at least 50 participants in each state;
we stopped recruitment when the sample size per state reached
150 participants. Unsurprisingly, the number of MTurk respon-
dents in our sample was proportional to the number of people in
a state (linear regression of state population size predicting
number of MTurk respondents, with robust SEs: coefficient =
To ensure participants recruited on MTurk were paying at-
tention throughout the survey, we included an attention check, as
Jachimowicz et al. www.pnas.org/cgi/content/short/1617395114 1of5
commonly done on MTurk. Ninety-one percent of 5,721 partici-
pants passed the attention check. Our main analysis focuses on
participants who passed the attention check. However, when we
include participants who failed the attention check, results are
Participants responded to the same 13-item community trust
survey as above (α=0.84). To further disassociate community
trust from generalized trust, we also collected participants’re-
sponses to the same measure of 6-item generalized trust as above
(α=0.89). Although participants in this study were not exclu-
sively low income, we are able to generate state-level averages of
community trust because our previous findings indicate com-
munity trust does not differ with income. State-level variation in
community trust could be driven, for example, by population
density, health levels, crime rate, or unemployment rate.
We ran additional analyses that relate community trust to a
range of other variables. Community trust is not significantly
related to population density, age, or savings. However, we do
find that community trust is marginally related to levels of un-
employment (r=−0.27, P=0.056), such that higher levels of
unemployment are related to lower levels of community trust.
In addition, community trust is also related to the percentage of
individuals who identify themselves as being “unbanked”(r=−0.44,
P=0.001) (60), such that higher levels of unbanked citizens were
related to reduced levels of community trust. In addition, we find
that savings are significantly related to payday loan use (b=−1.26,
SE =0.59, P=0.033), such that higher levels of savings are related
to lower levels of payday loan use.
We also conducted linear regressions with both community
trust and generalized trust as independent variables and payday
loan use as a dependent variable, and only community trust (β=
0.143, SE =0.05, P=0.0076), but not generalized trust
(β=−0.007, SE =0.006, P=0.90), was a significant predictor.
Finally, we ran additional analyses on our data, controlling for
state-level differences in payday loan regulations. We accessed a
database (61) that categorizes the legislation on payday loan use
into “restrictive,”“permissive,”and “hybrid.”We reran analyses
for states where legislation is more permissive (n=3,052) and
find no significant differences in our results.
Study 3: Experimental Evidence
We recruited 120 participants (45% female, mean age =33.2 y,
SD =10.24) from MTurk. Participants were randomly allocated
to one cell in a 2 ×2 design, manipulating levels of felt income
(low/high) and levels of felt community trust (low/high). To in-
duce varying levels of felt income, we used previously developed
and validated scenarios (2). Participants in both conditions were
presented with four different scenarios that asked them to make
decisions. The scenarios differed in the financial implications
they had; whereas participants in the high felt-income condition
were asked to imagine scenarios with relatively minor financial
implications, those in the low felt-income condition were asked
to imagine scenarios with more severe financial implications. For
example, participants were asked to imagine the following: “The
economy is going through difficult times; supposed your em-
ployer needs to make substantial budget cuts. Imagine a scenario
in which you received a 5% (15%) cut in your salary. Given your
situation, would you be able to maintain roughly your same
lifestyle under those new circumstances?”Participants were then
asked to elaborate, being prompted “Why or why not? If not,
what changes would you need to make? Would it impact your
leisure, housing, or travel plans?”[Other scenarios: Scenario 2:
Imagine that an unforeseen event requires of you an immediate
$200 ($2,000) expense. Scenario 3: Imagine that your car is
having some trouble and requires a $150 ($1,500) service. Un-
fortunately, your automobile insurance will cover only 10% of
this cost. Scenario 4: Suppose you have reached the point where
you must replace your old refrigerator. The model you plan to
buy offers two alternative financing options: (i) You can pay the
full amount in cash, which will cost you $399 ($999), or (ii) you
can pay in 12 monthly payments, of $40 ($100) each, which
would amount to a total of $480 ($1,200).] Imagining more se-
vere financial implications has been shown to mirror feelings of
actually having lower income (2).
We manipulated levels of community trust by increasing the
salience of this construct in the minds of respondents (51). We
gave participants a definition of community trust (“the extent to
which you trust your community”). Next, we asked participants
to either list 2 (low) or 10 (high) examples where community
trust was justified. This manipulation differs from Schwarz
et al.’s (52) ease-of-retrieval concept, as participants did not
experience difficulties in coming up with reasons for when
community trust was justified. Participants in the 10-example
condition wrote an average of 944.96 characters (SE =55.39),
more than three times as much than participants in the
2-example condition who wrote an average of 307.4 characters
(SE =20.43). Crucially, for participants in the 10-example condi-
tion, there is no difference in the number of characters written for
the first example (M=88.04, SE =5.71) relative to the last example
(M=104.28, SE =9.25; t=−1.49; P=0.139), indicating that
participants did not find it difficult to generate 10 examples. The
alternative prediction, based on research investigating the “ease-of-
retrieval”effect (52), would have predicted that participants would
find 10 examples harder to generate, such that the latter examples
were shorter. This would have led to lower levels of felt community
trust, which is opposite to what we find. Next, we assessed temporal
discounting factors using DEEP (43), similar to study 1. Finally, we
measured demographic control variables.
Study 4: Field Evidence from Bangladesh
In study 4, we sought to replicate our findings in a different
context with explicitly ultrapoor individuals. Given that our hy-
potheses target the future-oriented behavior of low-income
persons, it is imperative our findings from previous studies also
hold at the very low end of the income scale.
Before the start of the intervention, we surveyed 111 partici-
pants (37 females; mean age =38.3 y, SD =11.05) in 12 unions in
Bangladesh. (Age information was available only for 61 partici-
pants. According to the BRAC research team, this is a rough
representation of the overall age range of all participants.) Al-
though there are no specific records available, the BRAC re-
search team estimates that on average, most respondents are
educated up to grade 8 or lower. To highlight the extremely low-
income profile of our participants, we divided monthly house-
hold income by number of household members. On average,
surveyed participants indicated they had about 1,600 Taka (Tk.
1,600) for each household member (M=1603.53, SD =712.43),
which equals roughly $20.50. (Exchange rate Tk. 1,000 =$12.49,
February 13, 2017.) That is, household members had on average
less than $1/d. This is low even by Bangladeshi standards: Av-
erage monthly household income (M=7,208.56, SD =3,020.41)
is lower than average monthly rural household income (Bangladeshi
Taka) as measured by the World Bank Household and Expenditure
Survey in 2010 (Tk. 11,480). We assessed levels of temporal
discounting and community trust for this small sample of 111 par-
ticipants and found community trust to be a significant predictor of
temporal discounting (β=0.158, SE =0.037, P<0.001), such that
higher levels of community trust were associated with lower tem-
Following the intervention, interviewers approached partici-
pants individually and read aloud the questions. Following par-
ticipants’verbal responses, interviewers filled out the survey.
Levels of community trust were measured by adapting the 13-
item scale of study 2 to the local Bangladeshi context. After back
translation and pretesting at the local BRAC offices, we de-
veloped a 12-item scale equivalent for use in rural Bangladesh
Jachimowicz et al. www.pnas.org/cgi/content/short/1617395114 2of5
(α=0.69). Participants were asked to respond on a scale ranging
from 1 “not at all”to 5 “completely,”with sample items such as
“Daily life in my village makes me hopeful about the future of
my child(ren)”and “There is a strong sense of ‘community’and
‘trust’among the inhabitants of my village.”
Given contextual constraints, temporal discount factors were
assessed using a titration measure. Similar to previous research
(53), participants were asked to decide between a smaller–sooner
and a larger–later option. Interviewers explained the measure as
follows: “Listed below are several financial payout options
ranging from an instant payment of Tk. 500 to several different
incremental payout options available in 3 months’time. Please
indicate your payout preference for each row.”There were seven
payout options in 3 mo, ranging from Tk. 515 to Tk. 2,500 (the
local currency in Bangladesh; amounts were pretested by the
local BRAC office). For each option, participants were asked to
choose between Tk. 500 and the payout option in 3 mo. We were
unable to collect data at the end of the intervention for 4 of the
121 village union councils, 2 in the control and 2 in the treatment
First, we computed estimates of temporal discount factors by
calculating the indifference point, that is, the trade-off where
individuals switched from smaller–sooner to larger–later. There
were no preference reversals, for example participants preferring
Tk. 600 in 3 mo over an instant Tk. 500, but then preferring an
instant Tk. 500 over Tk. 700 in 3 mo. Sixteen participants always
chose the instant Tk. 500, even when the payout in 3 mo was Tk.
3,000. We code these participants as having an indifference point
of Tk. 3,001, a conservative estimate given that their true in-
difference point is unknown and possibly much higher. (There is
also the possibility that these participants did not fully un-
derstand the instructions given. Therefore, we ran all subsequent
analyses both with and without these 16 participants, with no
significant changes in our results.) On average, the indifference
point was M=1,851.08 (SD =1,041.93).
We also investigated whether individuals who were relatively
better off were less likely to be beneficially affected by higher
levels of community trust. To do this, we ran a two-way interaction
between treatment and average income on temporal discounting
rates. We find no significant interaction (P=0.94) and only a
significant main effect of treatment (P=0.021) on intertemporal
discount rate. However, we want to highlight the extremely low-
income levels of participants in this study. At the mean, partic-
ipants had less than $1/d per household member (∼$23.20/mo).
At 1 SD above the mean, participants had ∼$1.35/d (∼$40.64/mo),
and even at 2 SD above the mean, participants just barely sur-
passed the World Bank’s international poverty line of $1.90/d (at
∼$1.94/d or $58.11/mo). Thus, it is highly unlikely that the effect
of community trust should vary for relatively richer participants
in our study because most, if not all, participants in our sample
had trouble making ends meet and were thus likely beneficially
affected by higher levels of community trust.
To alleviate concerns of endogeneity we also ran a two-stage
least-squares model (62), with treatment as the instrumental
variable, community trust as the explanatory variable, and dis-
count factor as the dependent variable. The model results are
consistent with the hierarchical linear model specified above, such
that community trust remains a significant predictor of discount
factor (β=0.27, SE =0.13, P=0.032). Gender, the only de-
mographic variable that differed across conditions, had no sig-
nificant effect on temporal discounting, similar to findings of
prior research (63).
We also conducted 42 qualitative interviews and 8 focus group
discussions in 14 union councils, 7 treatment and 7 control. These
were conducted at the end of the intervention period by one of the
authors. Interviews were held with elected local government
representatives (24 individuals) and persons involved in the
community trust intervention (18 individuals). The focus group
discussions were held with local community members.
In the control unions, the lack of trust in one’s local community
is exemplified in quotes from interviewees, such as the following:
“Some of us attended a ward shava [local government meeting]
once, and we did not know what we were supposed to do. We
raised some concerns in front of the Member [local government
representative], but later nothing was done about it. It seems our
opinions do not matter, and so people from our community do
not attend these events or bother to find out what is happening
in the union.”“What is the point of going to meetings? We do
not have any say in this process. [...] There is no reason for us to
attend.”“The community does not get involved in my problems.”
“There is little to no participation from the local community.
The people who attend [local government meetings] are a lim-
ited number of influential people from the area.”“We hold
meetings regularly as we are supposed to, and try to get the
community involved, but the truth is people are simply not in-
terested. It is usually us representatives, local influential people
and local government officials who attend.”
In contrast, individuals in the treatment councils reported the
following: “When I fail to do something, they [other community
members] help us give the solution of the problem. Thus, we
work together.”“The people who got training try to arrange
meetings with regular citizens regularly which are named
‘Utahan Baithak’[courtyard meetings]. [...] The meetings are
held in different areas every week. The problems of the people
and their solutions are discussed here in this meeting.”“If you
need us while applying for this [government program], we will be
there to help you.”“After participating in the training, she calls a
weekly meeting at Friday in the yard of her home. [...] In these
gatherings, the community gets together to collectively discuss
and solve their problems. We have also formed a citizen’s
committee to find out the problems of the population and also
the solutions of these problems.”More details on the qualitative
data collected are available in a separate document, ref. 64.
Jachimowicz et al. www.pnas.org/cgi/content/short/1617395114 3of5
Low Community Trust High Community Trust Low Community Trust High Community Trust
Low Income High Income
Temporal Discounting Factor
Fig. S1. Temporal discounting factor as a function of income and community trust.
Table S1. Temporal discounting predicted by income and community trust
Model 1 Model 2 Model 3 Model 4
(Intercept) 0.1269*** 0.0913*** −0.0053 −0.0127
(0.0089) (0.0243) (0.0484) (0.0523)
Income 0.0034* 0.0221* 0.0204*
(0.0015) (0.0088) (0.0089)
Community trust 0.0149* 0.0376** 0.0371**
(0.0067) (0.0135) (0.0135)
0.0082 0.0077 0.0203 0.0257
0.0067 0.0062 0.0157 0.0166
No. observed 647 647 647 647
rmse 0.1121 0.1121 0.1116 0.1116
*P<0.05, **P<0.01, ***P<0.001.
Table S2. Field study descriptive statistics
No. of union councils 60 61
Community trust 3.31 (0.60) 3.45 (0.58)
Generalized trust 3.67 (0.71) 3.70 (0.71)
Mean age, y 33.83 (10.5) 33.10 (11.09)
Sex: 0 =male, 1 =female 0.31 (0.50) 0.47 (0.47)
Average monthly income,
in Bangladeshi Taka
1,935.31 (1,367.34) 1,956.94 (1,525.75)
Education 10.80 (2.95) 10.97 (2.72)
SD is in parentheses. Average monthly income is reported as household income divided by house-
hold members. One thousand Taka =$12.49 (February 13, 2017). Chi-square test reveals only gender
differed between conditions; otherwise there were no significant differences.
Jachimowicz et al. www.pnas.org/cgi/content/short/1617395114 4of5
Table S3. Hierarchical linear regression with unions nested
within condition and dependent variable temporal
Model 1 Model 2 Model 3
(Intercept) 0.39*** 0.10 0.06
(0.06) (0.09) (0.11)
Treatment: 0 =control,
0.08* 0.09* 0.08*
(0.03) (0.03) (0.03)
Generalized trust 0.54*** 0.54***
Sex: 0 =male, 1 =female 0.01
Average income 0.00
AIC −49.04 −62.79 −12.53
BIC −29.72 −38.64 30.90
Log likelihood 28.52 36.39 15.26
No. observed 928 928 928
No. groups 117 117 117
*P<0.05, ***P<0.001. AIC, Akaike’s information criterion; BIC, Bayesian
Jachimowicz et al. www.pnas.org/cgi/content/short/1617395114 5of5