NBER WORKING PAPER SERIES
TOWARD AN UNDERSTANDING OF THE DEVELOPMENT OF TIME PREFERENCES:
EVIDENCE FROM FIELD EXPERIMENTS
James Andreoni
Michael A. Kuhn
John A. List
Anya Samek
Kevin Sokal
Charles Sprenger
Working Paper 25590
http://www.nber.org/papers/w25590
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
February 2019
We thank the Kenneth and Anne Griffin Foundation and the National Institutes of Health (NIH)
grant
5R01DK114238 for funding this project. Andreoni also acknowledges the financial support
from the
National Science Foundation, Grant SES-165895. We thank participants at the AEA
meetings and
Nadia Chernyak for helpful comments. We thank the directors, principals and staff
at the Chicago
Heights Early Childhood Center and Illinois School District 170 for
accommodating the data collection
process. We thank Edie Dobrez, Jennie Huang, Phuong Ta,
Kristin Troutman, Andre Gray and our
staff of assessors for valuable research assistance. This
research was conducted with the approval of
the University of Chicago and University of
Southern California Institutional Review Boards. The
views expressed herein are those of the
authors and do not necessarily reflect the views of the National
Bureau of Economic Research.
NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official
NBER publications.
© 2019 by James Andreoni, Michael A. Kuhn, John A. List, Anya Samek, Kevin Sokal, and
Charles
Sprenger. All rights reserved. Short sections of text, not to exceed two paragraphs, may
be quoted
without explicit permission provided that full credit, including © notice, is given to the
source.
Toward an Understanding of the Development of Time Preferences: Evidence from Field
Experiments
James Andreoni, Michael A. Kuhn, John A. List, Anya Samek, Kevin Sokal, and Charles
Sprenger
NBER Working Paper No. 25590
February 2019
JEL No. C9,C93,D03
ABSTRACT
Time preferences have been correlated with a range of life outcomes, yet little is known about
their
early development. We conduct a field experiment to elicit time preferences of over 1,200
children
ages 3-12, who make several intertemporal decisions. To shed light on how such
primitives form,
we explore various channels that might affect time preferences, from
background characteristics to
the causal impact of an early schooling program that we developed
and operated. Our results suggest
that time preferences evolve substantially during this period,
with younger children displaying more
impatience than older children. We also find a strong
association with race: black children, relative
to white or Hispanic children, are more impatient.
Finally, assignment to different schooling opportunities
is not significantly associated with child
time preferences.
James Andreoni
Department of Economics
University of California, San Diego
9500 Gilman Drive
La Jolla, CA 92093-0508
and NBER
andreoni@ucsd.edu
Michael A. Kuhn
Department of Economics
1285 University of Oregon
Eugene, CA 97403
mkuhn@uoregon.edu
John A. List
Department of Economics
University of Chicago
1126 East 59th
Chicago, IL 60637
and NBER
jlist@uchicago.edu
Anya Samek
Center for Economic and Social Research
University of Southern California
635 Downey Way
Los Angeles, CA 90089
anyasamek@gmail.com
Kevin Sokal
University of Chicago
sokal@uchicago.edu
Charles Sprenger
University of California, San Diego
Rady School of Business
9500 Gilman Drive
La Jolla, CA 93093
c.sprenger@gmail.com
2
1. Introduction
The rate of time preference as elicited in the laboratory is strongly associated
with a range
of life outcomes, including health status, educational attainment
, and labor market earnings
(Golsteyn et al., 2014).
1
Among children
and adolescents,
higher rates of impatience
have been
linked to
a greater number of disciplinary referrals at school, lower high school completion rates
,
and more money spent on alcohol and cigarettes (Castillo et al., 2011; 2015; Sutter et
al., 2013).
In addition, impatient children are more affected by incentives than their patient counterparts
(Oswald and Backes-Gellner, 2014).
2
Therefore, how
intertemporal preferences form at an early
age, and how they interact with the environment, have direct policy implications.
This paper makes three overarching contributions to our understanding of the development
of time preference. First, we design and implement a time preference elicitation task in which
children ages
3-12 years old make a
series
of choices
between receiving smaller amounts of candy
at the end of the day
or larger amounts of candy on the next day.
There is a growing literature
seeking to understand
how economic preferences, such as time preferences, form at an early age.
Yet the assessment of children’s
preferences is still in its infancy
and a consensus is yet to form
about best methods.
We simplify the
elicitation tasks
typically used with adults
and adjust the
incentives to make the measures
developmentally
appropriate and incentive-compatible for
the
children in our sample.
3
1
In related work a
mong adults, time preferences predict health, smoking, drinking and drug abuse behaviors (Bradford
et al., 2017
; Chabris et al., 2008; Khwaja et al., 2006; Weller et al., 2008), demand for medical screening tests or
vaccines (Picone et al., 2004; Chapman and Coups, 1999) and take up financial education programs (Meier and
Sprenger, 2013).
2
In a related paper, Cou
rtemanche et al. (2015) find that impatient adults are more sensitive to food price changes
and exhibit the largest weight gain when food prices fall.
3
Another advantage of our measure with children is that it might be a 'purer' measure of time preference
than most of
the literature presents with
adult subject pools. One can think of this contribution in terms of measuring risk posture.
Conventional expected utility theory recognizes the important effects of background risk on risk attitudes measured
on the current choice. Harrison et al. (2007) show that background risk is important empirically, in that they find their
subjects are considerably more risk averse when background risk is introduced. This result suggest
s the import of
understanding the com
plete portfolio of risk the agent holds when making their choices. Similar reasoning should
hold in standard models of time preference and their measurement. Provided that our subjects did not have material
3
Our second contribution is
exploring
the correlates of time preferences.
An advantage of
this paper relative to
prior
work
is that our dataset is very comprehensive. We have data on child
demographic background (age, gender, race) and household characteristics (parents’ educational
attainment, household income). We go beyond these basic variables to collect detailed data on
child cognitive and executive function
skills
via a rigorous skills assessment. We also collect data
on a sub-set of the children’s
parents, which allows us to evaluate whether child time preferences
are associated with their parents’ time preferences.
This
lends insights into the origins of time
preferences.
Related work has explored the association of parents and family background with risk
preferences (Alan et al., 2014), competitiveness (Almås et al., 2015) and other
-regarding
preferences (Bauer et al., 2014; Wilhelm et al., 2008).
We find that time preferences evolve significantly as children
age, with younger children
displaying more impatience than older children.
This is
in line with related work
that finds a similar
association with age (Bettinger and Slonim, 2007; Angerer et al., 2015; Deckers et al., 2015; Sutter
et al., 2015). We also find a strong association with race: black children are significantly more
impatient
than
white or Hispanic children, even while controlling for socio-economic status,
cognitive
skills
and executive function skills
. Only one other paper has
had been able
to explore
this race relationship, and
it found a similar association for adolescents (Castillo et al., 2011).
Studying the associations of time preferences with race is important
since
- given that time
preferences predict academic outcomes
- it may help us understand the origins of the academic
achievement gap.
We do not observe a correlation between
preferences of parents and their children
. We
might have expected
such
a correlation due to genetics or social learning.
However, the
results in
the related
literature on the inter-generational transfer of time preferences
are also mixed. Kosse
background temporal risk, they should not be su
bject to this issue. As far as we are aware
, however, the literature has
not provided estimates of how background temporal profiles affect current choices.
4
and Pfeiffer (2012, 2013)
do find associations of time preferences of preschool children and their
parents, while Bettinger and Slonim (
2006) do not find an association with
children ages 5-16 and
their parents.
Researchers also
find some support for a link between future orientation of parents
and young adult children (Webley and Nyhus, 2006; Brown and Van der Pol, 2015).
Understanding the associations of time preferences of parents and children is important in light of
the recent interest in investing in parents as
a policy tool for human capital accumulation
(Fryer et
al., 2015).
Our third contribution is to evaluate the causal influence of early childhood education on
child time preferences. For this
evaluation, we take advantage of the Chicago Heights Early
Childhood Center (CHECC) study (Fryer et al., 2015; 2018). Children in ou
r study are participants
in CHECC, which randomly assigned children and parents from Chicago Heights, Illinois and
surrounding areas to
1) a free, high-quality preschool program,
2) a parenting program in which
parents were taught how to implement componen
ts of the preschool curriculum at home, or
3) to
a control group that did not receive an intervention. We hypothesized that children
randomized
to
CHECC preschool might become more patient
since they were exposed
to an environment and
activities that promoted patience, such as a structured preschool day, turn
-taking and modeling
patience.
The parenting program at CHECC also provided tools that parents could use to teach
patience
- such as a unit on self-regulation
– hence, we hypothesized that children exposed to
CHECC parent programs might also become more patient than children in the control group.
Our evaluation of CHECC joins a very small literature aimed at studying the causal
impact
of education programs on time preferences. Alan and Ertac (2014) found
that random assignment
to a program aimed at helping
third and fourth
grade
children imagine their future selves increased
patience relative to children assigned to a control group.
Lührmann
et al. (2014) found
that random
assignment
of adolescents to a
financial education program increased time consistency relative to
5
those
assigned to a control group. Unlike these studies, our early childhood interventions do not
focus specifically on time preferences and are broader in scope.
We believe that it is impor
tant to
learn whether “standard” early childhood programs, designed to impact cognitive abilities, also
affect time preferences.
Moreover, we explore time preference development in very early
childhood, which is a critical period of non-cognitive skill development (Heckman, 2000).
Our
study also speaks to the literature that uses early childhood interventions to understand the impact
on the academic achievement gap, such as
High/Scope
Perry Preschool and the Abecedarian
project
(Schweinhart, 1993; Campbell et al., 2002).
The evaluation of High/Scope Perry and
Abecedarian did not consider time preferences as we do, and
the sample size of these programs
was significantly smaller than ours.
In contrast to our hypotheses, we do not
find a statistically significant impact of CHECC
programs on time preferences.
This is true both immediately after the intervention as well as a few
years after the end of the intervention.
By contrast,
another paper evaluating CHECC
found an
impact of the preschool and parent programs on fairness and efficiency concerns but not on
selfishness
(Cappelen et al., 2016).
The fact that our early interventions, which were quite broad,
did not lead to
durable
changes in time preferences suggests that such preferences may be difficult
to change with
education
programs for 3-5 year-olds.
An important caveat is that we have
substantial attrition in our analysis sample.
The population
we study is also policy relevant.
By virtue of being from CHECC, the
households in our sample are of generally low SES.
Understanding how time preferences form
may be even more important among low SES children, since they are the ones most likely to exhibit
impatience (Deckers et al.,
2015; S
childberg-Hörisch et al., 2014), and may therefore benefit the
most from policy interventions.
Eckel et al. (2010) note that results from undergraduate students
do not always generalize to children or other populations.
Finally,
our study includes a much
6
broader age range than most other papers (for example, Castillo
et al., 2011; Sutter et al., 2013
focus on adolescents, while Kosse and Pfeiffer, 2012; 2013
, Falk and Kosse, 2016
focus only on
preschoolers).
In what follows,
Section 2 discusses our time preference elicitation, summarizes our data
and provides a discussion of the strengths and weaknesses of our measure. Section 3 discusses the
correlates of time preferences, including age, race, and parent time preferences. Section 4 explains
CHECC in more detail and provides the causal evidence. Section 5 concludes.
2. Time Preference Elicitation
2.1 Experimental Design and Procedures
The experiment was conducted in 4 waves.
In the
first three
waves
of the experiment
(2010-
11, 2012 and 2013),
families brought their children to the CHECC center
outside of school time
to participate. Participants did not know what the experiments were about when they signed up,
and participation was voluntary. Participation took approximately 30 minutes and parents received
approximately $25 for their participation.
In the last wave (2017-18), we conducted the
experiments during school and children were pulled from class to participate individually. The
sessions
differed in their implementation
, as described below.
Most children participated 1-2 times
between 2010 and 2018.
The basic
experimental design of the time preference elicitation task followed a multiple
-
price list format with 3-4 decisions (Coller and Williams, 1999). Eliciting time preferences in this
way has been shown to be correlated with life outcomes of adolescents and a
dults (e.g., Castillo et
al., 2011). Children made a series of decisions in which they were asked to choose between a
smaller amount of
rewards
on the day of the experiment at the end of the day (“at the end of the
day TODAY”), and a larger amount of
rewards
on the day after the experiment (“at the end of the
7
day TOMORROW”). Only one of the decisions “counted” for payment, and this was randomly
selected at the end of the experiment.
4
In the earlier sessions,
rewards
from the relevant decision
for payment were placed in paper bags with the date of payment on them and were given to the
child’s parents with a note providing instructions for when to give the child the candies. We also
verbally explained to parents when to give the
rewards
to the child.
5
In the
2017-18
wave
that was
conducted during school time, we
gave bags of
rewards
to teachers
to put in child backpacks
on
the dates that children selected.
Table 1 summarizes the series of decisions in each experimental
session.
[ TABLE 1: CHILD EXPERIMENT DESIGN ]
For most children, the experiment was conducted one-on-one with a trained experimenter
and each decision was accompanied by physical containers holding the number of
rewards
that
would be earned by the child for each alternative.
The rewards were always candies in waves 2010-
11, 2012 and 2013; and were the choice of different candies or prizes in 2017
-18. Some of the
older children (ages 6-12) in
the 2010-11
wave participated in small groups whereby children
circled pictures of candies on their recor
d sheets in private while experimenters walked around to
assist. The age overlap in procedures allows us to control for differences in implementation
approach.
4
For children ages 3
-5, the random selection was done in the following way.
Children were told tha
t at the end of the
session, one of their decisions would be selected at random as the ‘decision that counts.’ The ‘decision that counts’
was selected by having the child close his or her eyes and select one of X containers in the bin, each of which held t
he
candy and time for the candy to be given to the child for one of the decisions.
For children ages 6
-12, the random
selection was done via bingo cage at the front of the experiment room.
5
The potential for parents to not follow through on the experiment
al timing, and the child’s expectation thereof,
presents a potential confound in our study. If parents are likely to give their children the candy as soon as possible,
children should choose the most candy possible and, hence, appear quite patient in our s
tudy. This prediction is in
contrast to aggregate behavior, which exhibits substantial impatience.
8
2.2 Data
Table 2 provides a summary of the observations in our dataset, by
data collection wave.
A
total of
1,265
individual children participated in our experiments, with
926 participating in only
one wave,
307 participating in two waves and
32 participating in 3
waves. This gives us a total of
1,636
observations, spanning ages
3 through
12 (Mean=6.95, S.D.=2.64).
About half the
observations
were girls (50.03%). In line with the population of Chicago Heights, IL, our sample
is highly diverse, with
35.16% black and
55.77% Hispanic
observations. The households are
relatively low income:
28.42% of
observations
come from a household with an annual income of
$0-$15,000 and
27.20% come from a household with an annual income of $16,000
-$35,000.
About
17% of the
observations have
mothers
who do not have
a high school diploma, while
35% have a
high school diploma or some college education and
22% have a college degree.
[ TABLE 2: SUMMARY OF OBSERVATIONS ]
Figure 1 provides a histogram of the proportion of patient decisions (giving up fewer
rewards
today to choose more
rewards
tomorrow) across all sessions. It is notable that a large
proportion (28.97%) of children always select the
earlier, smaller reward while a small proportion
(12.04%) always select the later, larger reward.
[ FIGURE 1: HISTOGRAM OF CHILD DECISIONS ]
We also find that a sizable fraction of the children exhibit non
-monotonicities in their
choices, preferring a larger,
later number of
rewards
to a smaller,
sooner number, and subsequently
preferring an even smaller,
sooner number of
rewards
to the aforementioned later
, larger number.
The overall proportion of children displaying such non
-monotonicities is
40.63%.
However,
68.87% of the
965 children who are not always impatient or always patient are non
-monotonic.
Despite the high frequency of non-monotonicities,
as displayed in Figure 2,
we do observe that in
the aggregate children are more likely to be patient whe
n the cost of being impatient is high (i.e.,
9
when the difference between the earlier and later rewards is largest), a finding that is also observed
in Lemmon and Moore (2007) for children aged 4-5.
[ FIGURE 2: PROPORTION PATIENT CHILDREN BY DECISION & WAVE ]
2.3 Discussion
Our time preference elicitation methodology is similar to that used with adults in
experimental economics, and is in line with related work in developmental psychology that uses
children as young as age 2-3 to study future orientation (Schwarz et al., 1983; Lemmon and Moore,
2007; Garon et al., 2012). Our elicitation is similar to Sutter et al. (2015), who conduct time
preference experiments with Kindergarteners and use
1 choice of
1 reward today versus
2 rewards
the next day. Different from Sutter et al. (2015), we used a series of questions with varying interest
rates rather than just one question.
Our elicitation is also similar to one of the elicitations in Angerer
et al. (2015), who include children ages 6-11 in their experiments and use a series of questions in
which children choose between 2 tokens (which can be exchanged for candy or prizes) at the end
of the experimental sessions versus 3, 4 or 5 tokens in 4 weeks.
Our elicitation is also similar to
Bettinger and
Slonim (2007), who include children as young as 5 in their experiments
, but the
series of choices is delayed further in time
– by 1-2 months rather than by
1 day as in our study.
We believe that the shorter delay is more appropriate, since in
developmental psychology, a
1-day
delay is sometimes considered a “long” delay condition
for this age group (Schwarz et al., 1983).
Since the
high degree of non-monotonicities
of the children will not allow us to calculate
or estimate a conventionally mean
ingful discount rate,
in our analysis
we use two non-parametric
measures of time preference. The first measure is the total number of patient decisions
(standardized by session). The second measure is a binary variable indicating whether a child is
always
impatient or not.
Despite the non-monotonicities, we
believe the
elicitation task
is still
10
useful
since
it allows us to
categorize children with narrower bracketing than a single question
measure.
A different method for eliciting the impatience level of y
oung children is Mischel’s
“marshmallow” paradigm (Mischel et al., 1972; Mischel and Moore, 1973
; Mischel et al., 1989).
In this experiment, preschool aged children are seated in front of a treat and are offered the option
to either eat the treat, or to w
ait to receive double the amount. This paradigm is commonly used in
the developmental psychology literature (e.g., Karniol et al., 2011) and was also used by Kosse
and Pfeiffer (2012, 2013) to study intergenerational transfer of impatience from mothers to
their
preschool-aged children.
Developmental psychologists use the marshmallow paradigm because
unlike the “choice” paradigm, it puts children in a situation where they must overcome their
frustration and inhibit their
desire to eat the treat in front of them for a prolonged period of time
(Shoda et al., 1990). In the choice paradigm, children view the reward only briefly before making
their decision, and therefore are not in a prolonged situation where they must exercise inhibitory
control.
In our study, we
used the choice
paradigm
as our primary measure
because we
believe
that the choice paradigm, and not the marshmallow paradigm, is most similar to the time preference
elicitations that economists are interested in with adults.
A subset of the younger children in our study also participated in the marshmallow
paradigm
at different points in time than the main experiment
(881
observations with
799 children,
mean age=4.79, min of 3.2 and max of 7.6). In different
waves, we gave children either 5, 8 or 15
minutes wait time before the experimenter returned and doubled their treat.
Castillo et al. (2018)
use the time preferences data we report on here, the marshmallow
paradigm and a number of other
measures
not reported here
to study
associations of skills at an early age
and demonstrate that the
marshmallow paradigm is not correlated with the choice paradigm.
They also show that the time
preferences measured at an early age using our paradigm are associated with disciplinary referrals
11
several years later.
In this paper,
in the proceeding sections
we use the marshmallow paradigm as
an alternative measure of impatience to study the robustness of our findings.
A concern when evaluating time preferences with either children or adults is th
at they are
confounded with risk preferences (Andreoni and Sprenger, 2012). Participants may choose an
immediate reward rather than delaying the reward because they are risk averse and prefer a certain
outcome. We
address
this in two ways. First, all of our sooner, smaller rewards have a front-end
delay since children receive them “at the end of the day today.”
This
helps
to equalize
any
perceived
risk across payments.
Second, we also
directly
elicit risk preferences during
the session,
and we control for risk preferences in our analyses.
The risk preference elicitation in the 2010
-11
wave features the choice of a number of pencils from a jar, whereby one of the pencils has a red
mark on the bottom. Children get to keep all
the pencils, unless one of the pencils has a red mark.
If any pencil has a red mark, children must return all the pencils. This elicitation is summarized in
greater detail in Andreoni et al. (20
09). The risk preference elicitation in the remaining waves
features a multiple price list of choices between smaller, certain rewards and the different
probabilities of winning larger rewards. This elicitation is summarized in greater detail in Castillo
et al. (2018).
3. Correlates with Time Preferences
3.1 Age-Related Changes
Figure
3 provides a histogram of the ages in our sample and Figure 4
provides the trends
of patient decisions and consistency with age. Using the
proportion of
patient decisions
as our
main
measure, we find a
slight
decline in patience from about 3 years old to 5 years old, and a
larger
increase in patience from 5 years old to 12 years old.
Figure 3 also graphs the proportion of
decisions that are “all
immediate” or “all
delayed.” About 25% of decisions among 3-year-olds are
12
“all immediate”, and this number increases to nearly 50% for 5
-year-olds
and drops to under 10%
for children age 9 and up. Only about 10-20% of decisions at any age are “all
delayed.”
Figure 3
also displays the proportion of decisions that are mon
otonic,
including only those decisions with
at least
one switch point. We see that for children who have at least one switch point, monotonicity
increases from about 20% of observations among 3 year olds to about 30% of observations among
12 year-olds.
6
[ FIGURE 3: HISTOGRAM OF AGES ]
[ FIGURE 4: PATIENT AND MONOTONIC DECISIONS, BY AGE
]
The standard errors in the proportion patient are largest at the extremes of our age range.
The standard errors are smaller in the center of the age
distribution, where we see a clear positive
relationship between age and patience that is statistically significant in regression analyses.
Interestingly, we see some indication that children become less patient from age 3 to 5. We
attribute this to the possibility that some 3 year-olds have not yet understood the concept of
“tomorrow.” These children might choose the preferred, larger reward and not anticipate that they
will have to wait for it.
An indication that 3 year-olds might have difficulty with predicting
the
future
is presented in Busby and Suddendorf (2005), who find that only 30% of 3 year
-olds and
60% of 4-5 year-olds were able to correctly predict events that would happen tomorrow.
A confound with studying the evolution of time preferences with age is that other variables
are also changing during this time. For instance,
there are increases in cognitive abilities during
this same time period. In our analysis, we can control for cognitive abilities, as measured by a
score on a reading, writing
and math assessment administered within a year of the
experiments.
We can also control for executive functions, as measured by an assessment of inhibitory control,
6
A similar plot of monotonicity that does include the “all now” or “all later” data results in a decrease in
monotonicity with age. That is partly because
many more young children prefer “all now” than older children.
13
working memory and attention shifting.
7
Finally, we can control for risk preferences, which
may
also change during this time period.
Table
3 provides
regressions with proportion of patient decisions (standardized by session
,
specifications 1-4) and immediate choices (binary
, specifications 5-8) as dependent variables,
using all the observations
and clustering at the individual level.
All specifications feature wave
year controls.
In specifications
(2) and (6)
we add socio-economic characteristics, in specification
s
(3) and (7)
we add controls for cognitive ability and executive
functions, and in specifications
(4)
and (8)
we add the risk preference control.
The coefficient on age (row 1) is
positive
(between
0.05
and 0.09) and statistically significant in specifications
1-2, and negative (between 0.02 and 0.03)
and statistically
significant in specifications 5
-6, providing support for
the age trend displayed in
Figure 3.
Appendix Table A.1 includes an age squared variable
and shows
a weaker correlation
between age and time preferences
. However, the marginal effects do suggest that the relationship
between age and the proportion of patient decisions in these specifications is predicted to be
negative until ages 4-6, and positive thereafter.
[ TABLE 3: PREDICTORS OF CHILD TIME PREFERENCES ]
Studying the cross-sectional variation
in time preferences is
important because
time
preferences are predictive of later life outcomes. But studying the evolution of time preferences
by age is itself interesting since children make decisions that affect their future selves (such as
choice to complete homework, or show up to school).
The age profile of children’s patience
illustrates the degree to which older children will disagree with the decisions their younger selves
7
For participants below second grade, the cognitive abilities are measured by four sub
-tests of the Woodcock-
Johnson
III and the Peabody Picture Vocabulary III test. The executive functions ar
e measured using Blair and Willoughby’s
tests of working memory, attention shifting and inhibitory control. More details about each test are provided in Castillo
et al. (2018), which goes into detail on each sub
-test. For participants in third grade and a
bove, cognitive abilities are
taken from the NWEA MAP test administered by the state of Illinois each year, which is a personalized assessment
that measures individual student growth using a cross
-grade scale. Executive functions are taken from a separate
ly
administered assessment using the working memory and executive function and attention sub
-tests of the NIH
Toolbox.
14
made.
Further, many interventions are geared at this age range, and underst
anding the impact of
these interventions on children may involve understanding
where they are in the evolution of their
time preferences. For example, the evolution of time preferences we see here may suggest that
younger children would do better with immediate incentives while older children may accept
delayed incentives as part of an intervention.
3.2 Correlates with Race
We
next consider associations between child demographic
and
socio-economic
characteristics on child time preferences.
We find that child
race plays a statistically significant
role in the level of patience. Black children make a higher proportion of impatient decisions and
are more likely to make all impatient dec
isions relative to Hispanic children (see all specifications
in Table
3 – coefficient estimates are between
-0.16 and
-0.27
in specifications 1-4, with p-values
< 0.05).
In Appendix Table A.2
we also include an interaction term
between
race and age. The
interaction terms for the time preferences outcome variable are not statistically significant in most
specifications,
suggesting that the associations with race are similar across all ages in our sample.
Our finding that black children are more impatient is
in line with Castillo et al. (2011), who find
that among 13-14 year-old children, black children are more impatient than non
-black children.
Our sample includes children of ages 3-12, showing that this heterogeneity appears at even
very
young ages.
3.3 Correlates with Parent Time Preferences
The parent experiment included 16 decisions from two multiple-price lists, where parents
chose between amounts of $6 to $20 earlier versus $20 later. For the first 8 decisions the earlier
time was today and the later time was 5 weeks from today, and for the remaining 8 decisions the
15
earlier time was 5 weeks from today and the later time was 10 weeks from today.
Only one decision
was randomly paid out.
The parent time preference experiments were carried out in two wa
ves: once in 2012, and
again in 2017-18.
A total of
643 adult caregivers completed the parent preference elicitation tasks
(262
in 2012 and
381 in 2017-18).
501 participated only once and
71 participated two times.
Using
the original CHECC registration data, we identified
444 (77.62%) as the mother,
91 (15.91%) as
the father, and
36 (6.29%) as
another caregiver (usually this is the grandmother or relative that
lives with the child). For parent time preferences, we simply calculate the proportion of patient
decisions out of 16
(a histogram of these outcomes is available as Appendix Figure
A.2).
In case
of households that had multiple parents participating, we averaged the time preferences of both
caregivers for the analysis.
Since only a sub-set of parents completed the voluntary questionnaire
on socio-economic status, and a different (smaller) s
ub-set participated in the voluntary time
preference experiments, we consider both variables in separate regressions.
Table
4 presents regression results
including
the controls for parent time preferences. We
do not find strong associations of parent time
preferences with child time preferences.
The
coefficients on “Parent Time Pref.” are small and even change signs across specifications, with all
p-values above 0.10. Note that in Table 4 we continue to see the effects of age and race that we
described in sub-sections 3.1 and 3.2.
As a robustness check, Appendix Table A.3
replicates this
regression using only mothers, finding qualitatively similar results (no effect of mother’s time
preferences, and continued effects of age and race as described in sub
-secti
ons 3.1 and 3.2).
While parent preferences do not predict child preferences, as shown in Appendix Table
A.4, which regresses demographic characteristics of the child on the
parents’
time preferences, we
find that parents of black children are significantl
y more impatient than parents of Hispanic or
16
white children. This is in line with the race result for children presented in sub-section 3.2, and
suggests a persistence of measured time preferences into adulthood.
[ TABLE 4: PREDICTORS WITH PARENT CONTROLS ]
4. Impact of Early Childhood Interventions
4.1 Experimental Design
Our participants were recruited from the Chicago Heights Early Childhood Center
(CHECC) program.
8
CHECC is a large-scale intervention study on the role of different early
education programs on schooling outcomes
of disadvantaged children
conducted in 2010-2014
(Fryer et al., 2015; 2018).
Households who participated in CHECC
originated from the
surrounding area of Chicago Heights, Illinois. Chicago Heights is an ethnically diverse (41%
African American, 34% Hispanic) and
generally
low-income area (29% of persons below poverty
level, $18,121 per capita money income).
9
To support recruiting efforts, CHECC
ran a local
marketing campaign each year, which included direct mailings, automated phone calls to families
with children enrolled in the district, and information booths at community events in and around
the district. Program information was also distrib
uted through district leadership staff in the school
districts, and administrative assistants at schools were encouraged to collect and submit
registration forms for CHECC.
The main goal of CHECC was to investigate the role of early childhood programs on
educational attainment; therefore, households who signed up for the program were randomized
each year
(during four years 2010-2013)
into one of several different treatment arms or to a control
8
CHECC was called the Griffin Early Childhood Center (GECC) between 2010 and 2012, and was renamed to
CHECC in 2012.
9
Data from the Uni
ted States Census
http://quickfacts.census.gov/qfd/states/17/1714026.html
17
group.
10
A different set of treatments was tested in 2010
and 2011 and another set was tested in
2012 and 2013. The treatments are described below:
•
Preschool-Literacy and Math (2010 and 2011): This was a free, full-day 9-month long
preschool program that used the
Literacy Express
curriculum combined with a math
component. The purpose of this curriculum was to teach academic skills like literacy and
math.
•
Preschool-Tools of the Mind (2010 and 2011): This was a free, full-day 9-month long
preschool program that used
Tools of the Mind
curriculum. The purpose of this
curriculum
was to teach executive functioning skills.
•
Parent Academy-Cash (2010 and 2011): This was a class that parents attended two times a
month to learn how to teach to their children at home. Parents received $100 in cash for
attending each class, and
earned additional cash rewards for completing homework
assignments and for their child’s performance on tests.
•
Parent Academy-College (2010 and 2011): This was a class that parents attended two times
a month to learn how to teach to their children at home. Parents received $100 in cash for
attending each class, and earned additional rewards for completing homework assignments
and for their child’s performance on tests. The additional rewards were deposited into an
account they could access for their child’
s college (or other vocational, post-secondary)
education.
•
Preschool-CogX (2012 and 2013):
This was a free, half-day preschool program with half-
day of child-care, for 9 months. It also included a class that parents attended two times a
10
The CHECC
randomization followed a blocked approach. In each randomization, matched groupings of children
were created
based on gender, race (white, Hispanic or black), and age (within ½ years). Then, each child in the
grouping
was randomly assigned
to a treatment or control group.
Children for whom matched groupings were not
created were placed in the control
group. In Fryer et al. (2018) only the matched pairs are used and the full sample is
used as a robustness test, but in our analysis here we use the full sample.
18
month to learn how
to scaffold their children’s learning at home. Parents received $50 in
cash for attending each class, but did not receive additional rewards. The curriculum used
was
CogX,
which combines aspects of literacy, math, and executive functions and was
developed by the PIs (Fryer et al., 2018).
•
Kinderprep (2012 and 2013):
This was a free, half-day preschool program during the two
months of summer prior to the start of Kindergarten. It also included a class that parents
attended two times a month to learn how to scaffold their children’s learning at home.
Parents received $50 in cash for attending each class, but did not receive additional
rewards. The curriculum used was
CogX,
which combines aspects of literacy, math, and
executive functions and was developed by th
e PIs (Fryer et al., 2018).
•
Control group (all years):
Children randomized to the control group did not receive any
educational programming from us. This group was referred to externally as the Family
Group and families were invited to family
parties several times a year to minimize attrition
.
They also received cash
incentives to participate in assessments.
Fryer et al. (2015) reports on the impact of the Parent Academy programs, while Fryer et
al. (2018) reports on the impact of Preschool-CogX and Kinderprep on cognitive skills and
executive functions. The authors find that Parent Academy
primarily
improved executive
functions, while Preschool-CogX and Kinderprep
primarily
improved cognitive skills.
The impact
on cognitive skills faded out several years after the end of the programs.
In this paper, to investigate the impact
of early education programs on time preferences,
we use data from the 2012,
2013
and 2017-18
data collection waves
since these were conducted
after
most
children had the chance to participate in CHECC education programs.
There are some
caveats with the sample selection. In 2012 and 2013, we invited parents to participate in sessions
19
by bringing children in during a non
-school time
and we did not attempt to
recruit the full sample.
Only
39.87%
(815
of 2044)
of children who
had participated in a CHECC program were part of
the time preference data collection (
31%
- 284 of 921in 2012 and
27.1%
- 440 of 1,625
in 2013).
This
includes
46.62% of the Parent group,
50.47% of the Preschool group and
43.62% of the
Control group.
We used a different strategy in the
2017-18
wave. Instead of relying on parents to bring in
their children, in 2017-18, we collected data from all children who were attending one of the 9
schools in Chicago Heights Illinois School District 170. Data was collected during school.
Therefore, by design we do not have data on children who were attending other school districts
during this time period (data is available for
26.99% or
647 of 2208
of children).
This includes
30.42% of the Parent group,
21.30% of the Preschool group and
29.64% of the Control group.
However, if we believe that children did not move in and out of district due to CHECC treatment
assignment
– which they would have had no reason to do
– then this attrition should not affect the
results of our experiment.
Figure
A.1 in the appendix provides a diagram that describes
how children flow through
the programs and the experimental waves. Table A.
5 in the appendix provides summary statistics
comparing participants in the 2012-13 waves to non-participants from CHECC who would have
been eligible, and participants in the 201
7-18
wave who were in District 170 with non-participants
from CHECC who were not in District 170. We find that in the 2012
-13 waves, experiment
participants were similar to non
-participants on
race, gender, mother’s education and pre
-assessed
cognitive ability
(all p-values>0.10),
and different from non-participants on
age, income and pre-
assessed non-cognitive ability
(all p-values<0.01). We find that in the 2017-18
wave, participants
were similar to non-participants on
age, gender and pre-assessed non-cognitive ability
and
different from non-participants on race, income and mother’s
education. More participants in
20
2017-18
were Hispanic than in the overall sample (p-value<0.01) and fewer were black than in the
overall sample (p-value<0.01). These latter differences may have been expected
because District
170 is located in an area with more Hispanic residents
relative to areas where the rest of the sample
resides, and the 2017-18 wave was limited to District 170 students
.
It is important to delineate how this paper relates to other papers that have been published
using the CHECC sample. Fryer et al. (2015) and Fryer et al. (2018) report on the impact of the
programs on cognitive abilities and executive functions. Andreoni et al. (2018) reports on the
evolution of risk preferences of CHECC children and of adolescents
who participated in
a separate
intervention program. Cappelen et al. (2016) reports on the impact of the CHECC programs on
fairness preferences.
Unlike Andreoni et al. (2018) and Cappelen et al. (2016), we consider the
impact of the programs on time preferences.
Castillo et al. (2018) considers the associations of risk
preferences, time preferences, social preferences, cognitive abilities and executive functions at an
early age and evaluates the impact of these skills on disciplinary referrals several years
later.
Castillo et al. (2018) only use the time preferences (and other skills) collected at the beginning of
the CHECC study, while in this paper we use all of the time preference measures collected
throughout the CHECC study to understand the evolution of time preferences across ages.
Several
related papers also use small sub-samples of CHECC students to understand parental cheating
behavior (Houser et al.,
2016), parental charitable giving (Ben
-Ner et al., 2017; Samek and
Sheremeta, 2017), child charitab
le giving behavior (List and Samak, 2013; and List et al., 2018;
Cowell et al., 2015; Cox et al., 2016), child competitiveness (Samak, 2013)
and parent food choice
(Sadoff and Samek, 2018). A paper has also been written about the test-retest reliability of
executive function measures (Willoughby et al., 201
7).
21
4.2 Treatment Effects
Tables
5 and 6 show the impact of being randomly assigned to one of our interventions on
time preferences
, whereby Table
5 uses the 2012-13 waves
of data and Table
6 uses the 2017-18
wave of data.
The dummy variable “Preschool Dummy”
refers to whether the child was
randomized to any of the
preschool programs
(including the Kinderprep program), while the
dummy variable “Parent Academy Dummy” refers
to whether the child was randomized to any of
the Parent Academies. In some specifications, we also control for SES and cognitive and executive
function abilities at
baseline (when children entered CHECC)
. In Appendix Tables
A.6 and A.7
we perform the same analysis but disaggregating the
Parent Academy and Preschool
variables into
each
of the
separate
curricula treatment arms
described in
Section 4.1.
To mitigate concerns of
differential attrition, in
Appendix Tables A.8
and A.9
we perform the same analysis but using
inverse probability weighting by age, gender and race.
We do not see a strong association with randomization to one of the programs on child
time preferences
(all coefficients small
– on the order of 0.01 and 0.03
-- and insignificant with
p>0.10), suggesting that perhaps time preferences are difficult to influence through general
education programs such as ours.
For example,
specification (5) in Tables 5 and 6 provides
treatment effects of the programs on the choice of “always no
w.” We see that Preschool results in
an insignificant
2% decrease in the probability of choosing “always now.” We see that Parent
Academy results in either an insignificant
2% increase (Table 5, 2012-13 waves) or
5% decrease
(Table 6, 2017-18
wave) in the
probability of choosing of “always now”. The standard errors on
these coefficients are 0.03
to 0.13. By contrast, being black relative to Hispanic is associated with
a 14% increase in
the probability of choosing “always now.” And, in the Cap
pelen et al. (2016)
experiment that evaluated the impact of CHECC programs on fairness, being assigned to Parent
22
Academy is associated with a 14% increase in the probability of choosing the efficient versus fair
allocation of resources.
Note that race, but not age,
continues to be associated with time preferences in Table 5
. We
speculate that age is not statistically
significant in Table 5 because the 2013 wave includes only
children ages 3-6
(a more narrow age range). We speculate that race is not statistically significant
in Table 6 because the racial composition in the 201
7-18 wave is predominately Hispanic since
we collected data in one particular school district.
[ TABLE 5: TREATMENT EFFECTS, 2013 WAVE ]
[ TABLE 6: TREATMENT EFFECTS, 2017-18 WAVE ]
5. Additional Analysis
5.1 Multiple Hypothesis Test Correction
Tables
5-6 imply 5 different hypotheses are being tested, i.e., that
time
preferences evolve
with age, and may differ when comparing boys and girls, black and white children, black and
Hispanic children, and Hispanic and white children. It is thus important to adjust
for the family-
wise error rate (e.g., see List et al., 2016). Holm
-Bonferroni
p
-value correction yields continued
statistical significance for the comparisons of black and Hispanic children in
columns 3-5
of Table
3, as well as specifications 7-8 in Table
5. The association of age with time preferences
remains
statistically significant in specifications 1,2 and 5 in Table 3.
11
11
The Bonferroni procedure involves dividing 0.05 by the number of tests (5) and then comparing
each calculated
p
-
value to the new
p
-value of 0.01. The Bonferroni
-Holm procedure is sequential and compares the rank of each
p
-value
to 0.05/(5
-rank+1). Both procedures yield qualitatively similar results in our case.
23
5.2 Robustness Test with Marshmallow Paradigm
We also investigate the robustness of our results using the wait time on the marshmallow
test as the outcome variable. In Table A.10
in the Appendix, we report on regressions that use the
total number of seconds waited
as a dependent variable, setting all wait times to 5 minutes for
children who waited longer in sessions where it was feasible.
We find results that are qualitatively
similar to the results that use the time preference variable as an outcome:
an increase of 1 year in
child age is associated with an increase in wait time of about 6
-7 seconds and black children tend
to wait 7-27 seconds less than Hispanic children (black children also wait less than white children
in some specifications), however, the results are not statistically significant. We also do not find
an association of parent time preferences or effects of the Presc
hool and Parent Academy
treatments.
6. Conclusion
Time preferences are associated with a range of life outcomes, including educational
attainment, health, and financial capability
. To shed light on the development of time preferences
in children, we conducted experiments to evaluate correlations of child time preferences with age,
race, and parental
time preferences. We also
explored
the impact of
an early childhood
education
program on time preferences.
We found that time preferences evolve significantly during ages 3
-12, with younger
children displaying more impatient preferences than older children. We also found a strong and
significant association with race:
black
children, relative to white or Hispanic children, are
significantly more impatient.
Parent time preferences are not good predictors of child time
preferences, but parents of black children are also more impatient than parents of white or Hispanic
children.
Interestingly, assignment to different schooling opportunities are not significantly
24
associated with our measures of child time preferences
. More work is needed to understand
the
emergence of these observed racial differences, which are present at an early age.
There are
certain
limitations
within our data.
First,
it is unclear whether the ability to wait
is increasing with age because time perceptions change with age (i.e., 1 day to a 3
-year old feels
“longer” than 1 day to a 12-year old) or whether the underlying time preference construct i
s
changing. To disentangle these differences, future research should explore how changing the time
delay affects willingness to wait by age.
Future research should also explore the test-retest
reliability of this measure.
Second, it is unclear
whether parent preferences are uncorrelated with child preferences,
whether the measures that we use are the most appropriate for observing this correlation
, or
whether the preferences of children are simply difficult to measure
. Our results are in line with
Bettinger
and Slonim (2007) who also found no correlation between adolescent and parent time
preferences, but
are at odds with Kosse and Pfeiffer (2012; 2013).
Notably, we found no
association
in parent and child time preference using two different meas
ures of time preferences:
the standard economic time preference elicitation task, and the
delay of gratification paradigm. We
also found no association when constraining our sample to mothers only, as Kosse and Pfeiffer
(2012; 2013) do.
An interesting extension would be to systematically use alternative tests of parent
preferences, such as a qualitative question with parents, to see if differences in methodology can
partly explain the mixed findings in this literature.
Third, because our experiment was not
initially
designed to disentangle the causal impact
of schooling on child time preferences, we only see a sub-set of children in our data who were also
part of the CHECC randomization. Hence, while we do not see statistically significant differences
in time preferences by treatment assignment,
this could be
due to a small sample size or
due to
sample selection. For instance, suppose that random assignment to a CHECC treatment group does
25
causally affect child time preferences, but there is differential atten
dance at the experimental
sessions based on child level of impatience, such that parents of more impatient control group
children are less likely to attend than parents of more impatient treatment group children. Such a
story would undermine our ability to
find treatment effects.
To address this, we conducted a wave
of data collection in 2017-18 that assessed children in school.
This allowed us to reach all of the
children within one participating district, independent of parental involvement.
But this wave
occurred several years after the intervention, when the potential effects of the intervention on time
preferences could have faded out. We believe that future work should continue to use exogenous
variation in early childhood environments to better unders
tand the causal impact of such variation
on time preference development.
Finally, another
possibility is that early childhood education treatments are causally related
to making mistakes in the decision task, which could result in inconsistent decisions. H
owever,
when we re-run
specification (4) from Tables
5-6 with a 0/1 measure for “consistency” as the
dependent variable,
we do not observe statistically significant coefficients on CHECC treatment
assignment. This is reported as Appendix Table A.11.
Taken together,
our results suggest
interesting racial disparities in time preferences
that
emerge from a very young age and appear to persist. A deeper understanding of the determinants
of these differences and the extent to which they can be influenced b
y interventions
are important
topics for future research.
26
6. References
Alan, Sule, and Seda Ertac. "Good things come to those who (are taught how to) wait: Results
from a randomized educational intervention on time preference."
Available at SSRN
2566405
(2014).
Alan, Sule, Nazli Baydar, Teodora Boneva, Seda Ertac, and Thomas F. Crossley. "Parental
Socialization Effort and the Intergenerational Transmission of Risk Preferences." (2014).
Almås, Ingvild, Alexander W. Cappelen, Kjell G. Salvanes, Erik Ø.
Sørensen, and Bertil
Tungodden. "Willingness to compete: Family matters."
Management Science
62, no. 8
(2015): 2149-2162.
Andreoni, James,
Amalia
di Girolamo, John
A. List,
Claire
Maccelvious, and Anya Samek, “Risk
Preferences Are Associated with Cognitive
Skills and Executive Functions in Children and
Adolescents,” working paper, 2018.
Andreoni, James, and Charles Sprenger. "Risk preferences are not time preferences."
American
Economic Review
102.7 (2012): 3357-76.
Andreoni, James and William Harbaugh, “Unexpected Utility and its Alternatives using Convex
Risk Budgets,” Working paper, 2009.
Angerer, Silvia, Philipp Lergetporer, Daniela Glätzle
-Rützler, and Matthias Sutter. "How to
measure time preferences in children: a comparison of two method
s."
Journal of the
Economic Science Association
1.2 (2015): 158-169.
Bauer, Michal, Julie Chytilová, and Barbara Pertold
-Gebicka. "Parental background and other
-
regarding preferences in children."
Experimental Economics
17.1 (2014): 24-46.
Ben-Ner,
Avner, John A. List, Louis Putterman, and Anya Samek. "Learned Generosity? An
Artefactual Field Experiment with Parents and Their Children."
Journal of Economic
Behavior & Organization
143 (2017): 28-44.
Bettinger, Eric, and Robert Slonim. "Using Experime
ntal Economics to Measure the Effects of a
Natural Educational Experiment on Altruism."
Journal of Public Economics
90, no. 8-9
(2006): 1625-648.
Bettinger, Eric, and Robert Slonim. "Patience among children."
Journal of Public Economics
91.1
(2007): 343-363.
Bradford, David, Charles Courtemanche, Garth Heutel, Patrick Mcalvanah, and Christopher
Ruhm. "Time Preferences and Consumer Behavior."
Journal of Risk and Uncertainty
55,
no. 2-3 (2017): 119-45.
Brown, Heather, and Marjon Van der Pol.
"Intergenerational Transfer of Time and Risk
Preferences."
Journal of Economic Psychology
49 (2015): 187-204.
Busby, Janie, and Thomas Suddendorf. "Recalling Yesterday and Predicting
Tomorrow."
Cognitive Development
20, no. 3 (2005): 362-72.
Campbell, F.A., C.T. Ramey, E. Pugnello, J. Sparling, and S. Miller
-Johnson. “Early childhood
education: Young adult outcomes from the Abecedarian Project.
”
Applied developmental
science
, 6.1 (2002): 42-57.
Cappelen, Alexander W., John A. List, Anya Samek, and Bertil Tungodden.
The effect of early
education on social preferences
. No. w22898. National Bureau of Economic Research,
2016.
27
Castillo, Marco,
Paul J. Ferraro, Jeffrey L. Jordan, and Ragan Petrie. "The today and tomorrow of
kids: Time
preferences and educational outcomes of children."
Journal of Public
Economics
95.11 (2011): 1377-1385.
Castillo, Marco, Jeffrey L. Jordan, and Ragan Petrie. "How much is patience worth? Discount rates
of children and high school completion," working paper, 2015.
Castillo, Marco, John
A. List, Ragan
Petrie,
and Anya Samek. “Drivers of Behavior at an Early
Age: Evidence from a Large Field Experiment,” working paper, 2018.
Chabris, Christopher F., David Laibson, Carrie Morris, Jonathon Schuldt, and Dmitry Tau
binsky.
"Individual laboratory-measured discount rates predict field behavior."
Journal of Risk and
Uncertainty
37.2-3 (2008): 237-269.
Chapman, Gretchen B., and Elliot J. Coups. "Time Preferences and Preventive Health
Behavior."
Medical Decision Making
19, no. 3 (1999): 307-14.
Coller, Maribeth, and Melonie B. Williams. “Eliciting Individual Discount Rates.”
Experimental
Economics
2.2 (1999): 107-127.
Courtemanche, Charles, Garth Heutel, and Patrick McAlvanah.
"Impatience, Incentives, and
Obesity."
The Economic Journal
125, no. 582 (2015): 1-31.
Cowell, Jason M., Anya Samek, John List, and Jean Decety. "The Curious Relation between
Theory of Mind and Sharing in Preschool Age Children."
Plos One
10, no. 2 (2015).
Cox, James C., John A. List, Michael Price, Vjollca Sadiraj, and Anya Samek.
Moral costs and
rational choice: Theory and experimental evidence
. No. w22234. National Bureau of
Economic Research, 2016.
Deckers, Thomas, Armin Falk, Fabian K
osse, and Hannah Schildberg-Hörisch. "How does socio-
economic status shape a child's personality?" (2015).
Eckel, Catherine C., Philip J. Grossman, Cathleen A. Johnson, Angela de Oliveira, Christian Rojas,
and Rick K. Wilson. "(Im) patience among Adolesce
nts: A Methodological Note." (2010).
Falk, Armin, and Fabian Kosse.
Early childhood environment, breastfeeding and the formation of
preferences
. No. 2016-036. 2016.
Fryer Jr, Roland G., Steven D. Levitt, and John A. List.
Parental incentives and early childhood
achievement: a field experiment in Chicago heights
. No. w21477. National Bureau of
Economic Research, 2015.
Fryer
Jr, Roland
G., Steven
D.
Levitt, John A.
List,
and Anya Samek. “Towards an Understanding
of What Works in Preschool Education,” working paper, 2018.
Garon, Nancy M., Julie Longard, Susan E. Bryson, and Chris Moore. "Making decisions about
now and later: Development of future-oriented self-control."
Cognitive Development
27,
no. 3 (2012): 314-322.
Golsteyn, Bart HH, Hans Grönqvist, and Lena Lindahl. "Adolescent time preferences predict
lifetime outcomes."
The Economic Journal
124.580 (2014): F739-F761.
Harrison, Glenn W., John A. List, and
Charles Towe. "Naturally occurring preferences and
exogenous laboratory experiments: A case study of risk aversion." Econometrica 75, no. 2
(2007): 433-458.
Heckman, James J. "Policies to foster human capital."
Research in Economics
54.1 (2000): 3-56.
28
Houser, Daniel, John List, Marco Piovesan, Anya Savikhin Samek, and Joachim Winter. "On the
Origins of Dishonesty: From Parents to Children."
European Economic Review
82
(2015):
242-54.
Karniol, Rachel, Lior Galili, Dafna Shtilerman, Reut Naim, Karin
Stern, Hadar Manjoch, and
Rotem Silverman. "Why superman can wait: Cognitive self
-transformation in the delay of
gratification paradigm."
Journal of Clinical Child & Adolescent Psychology
40.2 (2011):
307-317.
Khwaja, Ahmed, Frank Sloan, and Martin Salm. "Evidence on Preferences and Subjective Beliefs
of Risk Takers: The Case of Smokers."
International Journal of Industrial Organization
24, no. 4 (2006): 667-82.
Kosse, Fabian, and Friedhelm Pfeiffer. "Impatience among preschool children and their
mothers."
Economics Letters
115.3 (2012): 493-495.
Kosse, Fabian, and
Friedhelm Pfeiffer.
“Quasi-hyperbolic time preferences and their
intergenerational transmission.”
Applied Economics Letters
, 20.10 (2013): 983-986.
Lemmon, Karen, and Chris Moore. "The
development of prudence in the face of varying future
rewards."
Developmental Science
10.4 (2007): 502-511.
List, John A., and Anya C. Samak. "Exploring the Origins of Charitable Acts: Evidence from an
Artefactual Field Experiment with Young Children."
Economics Letters
118, no. 3 (2013):
431-34.
List, John A., Azeem M. Shaikh, and Yang Xu.
Multiple hypothesis testing in experimental
economics
. No. w21875. National Bureau of Economic Research, 2016.
List, John A., Anya Samek, and Dana L. Suskind.
"Combining Behavioral Economics and Field
Experiments to Reimagine Early Childhood Education."
Behavioural Public Policy
2, no.
01 (2018): 1-21.
Lührmann, Melanie, Marta Serra-Garcia, and Joachim K. Winter. "The impact of financial
education on adolescents' intertemporal choices." (2014).
Meier, Stephan, and Charles D. Sprenger. "Discounting Financial Literacy: Time Preferences and
Participation in Financial Education Programs."
Journal of Economic Behavior &
Organization
95 (2013): 159-74.
Mischel, Walter, Yuichi Shoda, and Monica L. Rodriguez. "Delay of gratification in
children."
Science
244.4907 (1989): 933-938.
Mischel, Walter, Ebbe B. Ebbesen, and Antonette Raskoff Zeiss. "Cognitive and Attentional
Mechanisms in Delay
of Gratification."
Journal of Personality and Social Psychology
21,
no. 2 (1972): 204-18.
Mischel, Walter, and Bert Moore. "Effects of Attention to Symbolically Presented Rewards on
Self-control."
Journal of Personality and Social Psychology
28, no. 2 (1973): 172-79.
Oswald, Yvonne, and Uschi Backes-Gellner. "Learning for a Bonus: How Financial Incentives
Interact with Preferences."
Journal of Public Economics
118 (2014): 52-61.
Picone, Gabriel, Frank Sloan, and Jr. Donald Taylor. "Effects of Risk
and Time Preference and
Expected Longevity on Demand for Medical Tests."
Journal of Risk and Uncertainty
28,
no. 1 (2004): 39-53.
Sadoff, Sally, and Anya Samek. "Can Interventions Affect Commitment Demand? A Field
Experiment on Food Choice."
Journal of Economic Behavior & Organization
, 2018.
29
Samak, Anya C. "Is There a Gender Gap in Preschoolers’ Competitiveness? An Experiment in the
U.S."
Journal of Economic Behavior & Organization
92 (2013): 22-31.
Samek, Anya, and Roman M. Sheremeta.
"Selective Recognition: How To Recognize Donors To
Increase Charitable Giving."
Economic Inquiry
55, no. 3 (2017): 1489-496.
Schildberg-Hörisch, Hannah, Thomas Deckers, Armin Falk, and Fabian Kosse. "How Does Socio
-
Economic Status Shape a Child's Personality?" (2014).
Schwarz, J. Conrad, Janet B. Schrager, and Andrea E. Lyons. "Delay of gratification by
preschoolers: Evidence for the validity of the choice paradigm."
Child
Development
(1983): 620-625.
Schweinhart, Lawrence J.
Significant
Benefits: The High/Scope Perry Preschool Study through
Age 27. Monographs of the High/Scope Educational Research Foundation, No. Ten
.
High/Scope Educational Research Foundation, 600 North River Street, Ypsilanti, MI
48198-2898 (1993).
Shoda, Yuichi, Walter
Mischel, and Philip K. Peake. "Predicting Adolescent Cognitive and Self
-
regulatory Competencies from Preschool Delay of Gratification: Identifying Diagnostic
Conditions."
Developmental Psychology
26, no. 6 (1990): 978-86.
Sutter, Matthias, Martin G.
Kocher, Daniela Glätzle-Rützler, and Stefan T. Trautmann.
"Impatience and Uncertainty: Experimental Decisions Predict Adolescents Field
Behavior."
American Economic Review
103, no. 1 (2013): 510-31.
Sutter, Matthias, Levent Yilmaz, and Manuela Oberauer. "
Delay of Gratification and the Role of
Defaults—An Experiment with Kindergarten Children."
Economics Letters
137 (2015):
21-24.
Webley, Paul, and Ellen K. Nyhus. "Parents’ influence on children’s future orientation and
saving."
Journal of Economic Psychology
27.1 (2006): 140-164.
Weller, Rosalyn E., Edwin W. Cook, Kathy B. Avsar, and James E. Cox. "Obese Women Show
Greater Delay Discounting than Healthy-weight Women."
Appetite
51, no. 3 (2008): 563-
69.
Wilhelm, Mark Ottoni, Eleanor
Brown, Patrick M. Rooney, and Richard Steinberg. "The
intergenerational transmission of generosity."
Journal of Public Economics
92, no. 10
(2008): 2146-2156.
Willoughby, Michael T., Laura J. Kuhn, Clancy B. Blair, Anya Samek, and John A. List. "The
Test–retest Reliability of the Latent Construct of Executive Function Depends on Whether
Tasks Are Represented as Formative or Reflective Indicators."
Child Neuropsychology
23
,
no. 7 (2017): 822-837.
1 Tables
Table 1: Child Experiment Design
Wave
Elicitation task
Incentives
Implementation
(today vs tomorrow)
2010-11 4v5, 4v6, 4v7, 4v8 Candies
One-on-one or in a group
2012
3v3, 2v3, 1v3
Candies
Outside of school
2013
2v3, 2v4, 2v5, 2v6 Candies
One-on-one, outside of school
2017-18 4v5, 4v6, 4v7, 4v8 Choice of Candies/Prizes One-on-one, in-school
(same as 2010-11)
Note: The table reports the experiment design for the child experiments, broken down by wave.
1
Table 2: Summary of Observations
Wave 2010-11 Wave 2012 Wave 2013 Wave 2017-18 Total
Child Age Range (Years):
3 - 12
4 - 8
3 - 6
6 - 12
3 - 12
Child Age (in Years)
5.60
5.24
4.73
9.76
6.93
(0.14)
(0.05)
(0.04)
(0.05)
(0.07)
Child Gender (Female=1)
0.48
0.48
0.54
0.49
0.50
Child Race - Black
0.37
0.51
0.41
0.23
0.35
Child Race - Hispanic
0.49
0.37
0.49
0.72
0.56
Child Race - Other
0.01
0.01
0.01
0.00
0.01
Child Race - White
0.13
0.11
0.09
0.04
0.08
Household Income (0-15k)
0.21
0.31
0.28
0.32
0.29
Household Income (16k-35k)
0.20
0.26
0.34
0.27
0.28
Household Income (36k-60k)
0.13
0.13
0.11
0.09
0.11
Household Income (60k+)
0.06
0.08
0.10
0.02
0.06
Mother Edu (Less than High School)
0.12
0.11
0.15
0.23
0.17
Mother Edu (High School)
0.31
0.40
0.35
0.35
0.35
Mother Edu (College)
0.20
0.33
0.31
0.13
0.23
Cog Pre-Assess.
0.37
0.39
0.37
0.29
0.34
(0.01)
(0.01)
(0.01)
(0.01)
(0.01)
Non-Cog Pre-Assess.
0.66
0.58
0.51
0.50
0.55
(0.01)
(0.01)
(0.01)
(0.01)
(0.01)
Pretest Cog Missing
0.04
0.13
0.09
0.20
0.13
Pretest EF Missing
0.04
0.13
0.11
0.22
0.14
Cog Missing
0.30
0.43
0.55
0.01
0.28
EF Missing
0.30
0.43
0.55
0.03
0.29
Income Missing
0.40
0.22
0.18
0.30
0.27
Mother Educ Missing
0.38
0.16
0.18
0.28
0.25
Observations
248
286
447
633
1614
Note: The table reports sample averages. Standard errors are in parentheses. The number of observations is the number of
subjects in each wave, regardless of if they participated in the previous wave. Total observations represents total number of
assessments conducted, rather than total number of children. Demographic data is available for nearly all observations. Age
is available for all observations. Gender is available for all but 7 observations (6 children), and race is available for all but 15
observations (15 children). SES data is only available for children whose parents completed the voluntary questionnaire.
2
Table 3: Predictors of Child Time Preferences
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Time Pref. Time Pref. Time Pref. Time Pref. Always Now Always Now Always Now Always Now
Child Age (in Years)
0.08
⇤⇤⇤
0.07
⇤⇤⇤
0.03
0.05
-0.03
⇤⇤⇤
-0.02
⇤⇤
-0.04
-0.04
(0.02)
(0.02)
(0.07)
(0.07)
(0.01)
(0.01)
(0.03)
(0.03)
Child Gender (Female=1)
0.06
0.06
0.09
0.12
-0.03
-0.03
-0.04
-0.05
(0.05)
(0.05)
(0.07)
(0.07)
(0.02)
(0.02)
(0.03)
(0.03)
Child Race - White
-0.02
0.01
0.00
-0.04
0.01
0.01
-0.01
0.01
(0.10)
(0.11)
(0.16)
(0.16)
(0.04)
(0.04)
(0.07)
(0.07)
Child Race - Black
-0.16
⇤⇤
-0.16
⇤⇤
-0.25
⇤⇤
-0.26
⇤⇤⇤
0.09
⇤⇤⇤
0.08
⇤⇤
0.11
⇤⇤
0.11
⇤⇤
(0.05)
(0.06)
(0.08)
(0.08)
(0.02)
(0.03)
(0.04)
(0.04)
Child Race - Other
-0.45
-0.45
-0.46
-0.44
0.25
0.27
0.34
0.34
(0.37)
(0.37)
(0.57)
(0.57)
(0.16)
(0.16)
(0.21)
(0.21)
R2
0.02
0.02
0.04
0.05
0.14
0.14
0.14
0.15
Test Black=White p-value
0.17
0.10
0.11
0.17
0.10
0.09
0.11
0.15
N
1614
1614
820
803
1614
1614
820
803
This table reports OLS coe