Fickle Groups: An Empirical Assessment of Time ......I am grateful to Michael Callen (Harvard Kennedy School), Aprajit Mahajan (UC Berkeley), and Charlie Sprenger (UC San Diego) for

Fickle Groups: An Empirical Assessment of Time

Preferences Using Experimental Data∗

Karrar Hussain †

USC

This Version: October 10, 2016

Very Preliminary. Please do not cite or circulate.

Abstract

Inconsistencies in individual intertemporal choices have been well documented in Eco-

nomics. This paper studies individual and collective decisions through the preference

elicitation method over task (real consumption). For this purpose, a joint experimental

elicitation of time preferences was performed for the groups (who have to reach a con-

sensus) as well as for their individual members. Using an analysis of the elicited choices

of decision making over time, the study found that group/joint decision making resulted

in much higher elicited present-bias (procrastination) as compared to individual decision

making. This finding is robust to a variety of alternative specifications. The results also

suggest that the group/joint aspect of the decision-making process exacerbates standard

measures of present-bias due to group members’ discount rate heterogeneity. The study

also concludes that the presence of a dominant individual based on bargaining power

improves the procrastination tendency in group.

JEL classification: C91, D12, D81

Keywords : Time Discounting, Preference Elicitation, Present-bias

∗I am grateful to Michael Callen (Harvard Kennedy School), Aprajit Mahajan (UC Berkeley), and CharlieSprenger (UC San Diego) for their advise and support. I am grateful to the Center for Economic ResearchPakistan (CERP), Sohaib Athar, Zara Salman, Mehroz Alvi (IGC, Pakistan) for excellent logistical support. Iam indebted to Dr. Turab Hussain and Muhammad Hussain (Lahore University of Management Sciences) forchampioning this project. Rao Hashim and Danish Raza provided excellent research assistance. (Research isapproved by the Institutional Review Board at Harvard and at UC San Diego.)†University of Southern California, 3620 South Vermont Avenue, KAP 300 Los Angeles, California 90089.

[email protected]

Draft—Not Ready For Circulation 2

1 Introduction

Intertemporal choices – decisions in which costs and benefits are spread out over time – are

both common and important aspects of life. Understanding the factors and dynamics associated

with these intertemporal choices is a valuable and important goal for any decision maker (policy

maker, household head, and managers etc.). In economics, the notion of intertemporal choices is

relevant in a wide range of issues such as consumption and savings, labor-leisure choice, building

reputations, education and health care, and investment decisions, because an economic agent’s

choice often entails tradeoffs through time. This is why depicting such choices with structural

models of discounting has occupied the interest of economists for much of the last century

(leading contributions include Samuelson, 1937; Koopmans, 1960; Laibson, 1997; O’Donoghue

and Rabin, 2001).

The preference parameters – intertemporal in this case – governing these decisions and

associated models are of unique value in understanding a broad range of behaviors and have

received much empirical attention. In both field and laboratory settings substantial efforts

have been made to structurally estimate the level and shape of discounting (examples include

Hausman, 1979; Lawrance, 1991; Warner, Pleeter, and Cagetti, 2003; Laibson, Repetto, and

Tobacman, 2005; Harrison, Lau, and Williams, 2002; Anderson, Harrison, Lau, and Rutstrom,

2008; Shapiro, 2005; Kuhn, 2013; Andreoni and Sprenger, 2012).

Research efforts (empirical and theoretical) have largely focused on modeling the sources

of time inconsistency at the level of the individual. This paper takes a different approach. It

considers how acknowledging the collective nature of choice can rationalize time inconsistencies

in revealed preferences settings. Many dimensions of intertemporal choice are better modeled as

an outcome of group, rather than individual, decision making. For instance, savings, education,

and health decisions are typically made at the collective household level, whereas, allocation

of budgets over time within firms and committees are products of multi-party deliberations.

Even in the context of individual choice, one can consider the existence of multiple selves with

distinct personalities rather than a single homogeneous decision making unit.1

Acknowledgement of the collective nature of choice, and comparing it with its individual

members’ choices, can help to rationalize the apparent time consistency or inconsistency in

the decision-making process. Theoretically, in a group context, inconsistencies can arise sim-

ply from the aggregation of heterogeneous preferences. Variations in individual discount rates

(Marglin,1963; Feldstein, 1964; Jackson and Yariv, 2015; Zuber, 2010) and innovations in the

Pareto weight summarizing the collective decision-making process, represent relevant consider-

1Shefrin and Thaler (1981) contrasts the long-sighted “planner within us to the short-sighted “doer, whileMetcalfe and Mischel (1999) contrast our hot and cool systems. Such evidence supports the application ofcollective choice models to characterize the behavior of individuals.


ations in this regard. In fact, Jackson and Yariv (2015) show that for a uniform distribution

of discount rates in an otherwise homogeneous population, group utility maximization in a

non-dictatorial way generates aggregate behavior that corresponds to hyperbolic discounting.

As a result, if other things remain equal, it is optimal to favor impatient group members of the

group in early periods and patient members in later periods.

There is little empirical evidence on group decision making as an intertemporal choice. Even

so, numerous theoretical papers are devoted to group decision making and the aggregation of

time preferences. Under that aggregation, these papers predict that a collective decision process

will most likely result in inconsistent choices over time, even if group members are individually

consistent (Gollier and Zeckhauser, 2005; Jackson and Yariv, 2015). Empirical evidence on the

aggregation of time preferences supports this view. For instance, Jackson and Yariv (2014)

show that a large majority of subjects acting as social planners are present-biased and that

only 2% of them exhibit time consistent behavior. Mazzocco (2008) showed that intertemporal

decisions of couples are different from singles decisions. Using the Consumer Expenditure

Survey, he showed that the traditional non-durable consumption smoothing behavior under

expected utility was satisfied for singles but not for couples. According to Mazzocco (2008),

this suggests that together with spouse preferences, the relative decision power of each partner

plays an important role in this asymmetry. A consequence is that by ignoring the change in the

balance of power, the revealed behavior of individuals in intertemporal decisions can be biased.

This paper contributes to the literature on this topic by gathering evidence that compares

groups and individuals in the domain of task consumption/effort allocation over time. In par-

ticular, we compare individual decisions with those of groups, whose members are matched

randomly. Since every participant makes the decisions individually and collectively, the order

in which these kinds of decisions take place becomes important, i.e. making the individual deci-

sions first, followed by the group decisions, or vice-versa. In the design part of the experiment,

we explicitly controlled for this order effect of decision making by selectively distributing the

ordering sequence among the groups.

The within-group parameters’ estimates generated in this paper yield new empirical evi-

dence on the outcome resulting from individual and collective decisions over time. The group,

along with its individual (two) members intertemporal allocation decision were used to test the

theoretical predictions of the time preference models. As described above using Jackson and

Yariv (2015), a set of testable hypotheses were generated and tested subsequently. For this

using the groups’ time preference estimates (specifically focusing on groups’ procrastination

tendency – related to present bias estimate) and their individual members’ structural estimates

(including discount rate, present bias, and effort cost parameters’ estimates), the roles and

effects of theory based heterogeneities are explored empirically on group procrastination ten-


dency. These within group heterogeneities include the differences in individual members’ (i)

discount factor, (ii) parameter governing the cost of effort, and (iii) bargaining position of each

person in the group.

In the sample of 244 participants, the study documented four principal results. First, on

aggregate, a present bias exists in participants individual behavior. This result is in line with

prior empirical and theoretical literature. Second, the degree of present-bias (procrastination)

was more pronounced in a group’s task allocation decisions as compared to an individuals

task allocation setting. This justifies the rationale for not forming the group for more or less

perfectly substitutable tasks. Third, the order in which decisions were made, whether making

the individual task allocation first and then the group task allocation or vice versa had no effect

on the degree of present-bias differences between individuals and groups. Lastly, using within-

groups estimates of present bias, the variations in group’s individual members’ discount rates

and bargaining power do explain group procrastination tendencies as postulated by Jackson

and Yariv (2015).

1.1 Literature Review

Following Samuelson (1937), and Fishburn and Rubinstein (1982), a large part of the theoretical

literature on time preferences builds on discounted utility and additively separable functional

forms that assume a separation between value and delay in assessing temporal sequences of

outcomes. A typical example is the exponential discounting utility model, which assumes

stationarity of time preferences and serves as the workhorse of many economic models. The

discounted utility model’s representation of time preferences also facilitates the use of empirical

measurements. With an extra assumption on the linearity of utility, measures of discount factors

and discount rates can be carried out by way of simple experiments (Thaler, 1981; Coller and

Williams, 1999). If one instead assumes nonlinear utility, then measurements become more

sophisticated but also more complex (Andreoni and Sprenger, 2012).

Notably large body of laboratory research has focused on identifying the shape of time

preferences in the context of time-date monetary payments (Recent papers using time-dated

monetary payments include Ashraf, Karlan and Yin, 2006; Andersen, Harrison, Lau and Rut-

strom, 2008; Dohmen, Falk, Huffman and Sunde, 2010; Tanaka, Camerer and Nguyen, 2010;

Benjamin, Choi and Strickland, 2010; Voors, Nillesen, Verwimp, Bulte, Lensink and van Soest,

2012; Bauer, Chytilova and Morduch, 2012; Sutter, Kocher, Glatzle-Ruetzler and Trautmann,

2013; and Dupas and Robinson, 2013). The empirical literature on time preference using a time-

dated monetary payments has elicited an extremely wide variety of discount rates. Frederick

et al. (2002) report elicited discount rates ranging from less than 1% (Thaler, 1981) to more

than 1,000% (Holcomb and Nelson, 1992). Moreover, individuals often exhibit present-bias and


thereby violate stationarity (Benzion et al., 1989; Kirby and Marakovic, 1995; Bleichrodt and

Johannesson, 2001; DellaVigna, 2009).

Experiments involving time-dated monetary payments have several confounds to identify

the shape of time preferences from such monetary choices. Issues of payment reliability and risk

preference suggest that subject responses may be closely linked to their assessment of the ex-

perimenter’s reliability rather than solely their time preferences. The main idea was originally

raised by Thaler (1981) who, when considering the possibility of using incentivized monetary

payments in intertemporal choice experiments noted ‘Real money experiments would be inter-

esting but seem to present enormous tactical problems. (Would subjects believe they would

get paid in five years?)’. Further empirical work validates Thaler’s (1981) suspicion. Andreoni

and Sprenger (2012a), Gine, Goldberg, Silverman and Yang (2010), and Andersen, Harrison,

Lau and Rutstrom (2012) all document that when closely controlling transactions costs and

payment reliability, dynamic inconsistency in choices over monetary payments is virtually elim-

inated on aggregate. Further, when payment risk is added in an experimentally controlled way,

non-expected utility risk preferences deliver behavior observationally equivalent to present bias

as described above (Andreoni and Sprenger, 2012b). Furthermore, monetary payments may not

be suitable to identify parameters of models defined over time-dated consumption. Arbitrage

arguments imply that choices over monetary payments should only reveal subjects’ borrowing

and lending opportunities (Cubitt and Read, 2007).

Evidence on group choice over time mainly concerns impatience. Available studies suggest

that groups are more patient than their individual members. For example, individuals are

more patient when making a joint decision with a partner than when making a decision for

themselves. This statement holds whether the group consists of a decision-making real-life

couple (Carlsson et al., 2012) or an experimental ‘artificial’ couple (Shapiro, 2010). Carlsson et

al. (2012) also find that couple-made decisions violate stationarity. For larger groups, collective

patience has been reported in groups of three to seven people (Shapiro, 2010; Denant-Boemont

and Loheac, 2011).

There are two key contributions of this paper. First using the domain of effort task con-

sumption, it elicits the time preferences of groups and their individual members simultaneously.

Using these elicited preferences it estimates the time preference parameters for both groups and

their individual members. Secondly, it connects the groups’ time preference parameter of time

inconsistency measures with theory based individual members’ characteristics. The paper pro-

ceeds as follows. Section 2 presents the setting of the experiment overview and provides a

theoretical background on time preferences. Section 3 describes the experimental design. Sec-

tion 4 summarizes the experimental results, and Section 5 concludes.


2 Experiment Overview

The duration of the experiment was three weeks in which 244 participants took part in the

study. All the participants were students from Lahore University of Management Sciences

(LUMS). The experiment was based on the premise of introducing high-resolution monitoring

technology. Each student in the sample was given a smart-phone equipped with a reporting

application. This application (discussed in detail in section 2) permitted precise observation of

when tasks were conducted, which in turn allowed for implementation of pay-for-performance

contracts. The task for participants was defined as taking pictures of pages of any book of their

choice using the high-resolution monitoring technology provided to them. However, this task

could only be completed in the official time window set for the different days of the experiment.

All participants were informed of this at the start of the experiment.

The intertemporal nature of multi-day task performance allowed for the introduction of

intertemporal bonus contracts. Participants were offered a fixed bonus of $15 for taking pictures

of either a total of V = 200 pages over two-days in the individual setting or a total of V = 400

pages over two days in the group setting, depending on which category they were assigned.

Each group consisted of two members matched randomly. The total number of pictures for

the first day was set as v1 and for the second day as v2. The experiment had 3 active days of

participation spaced out with a gap of a week between them. On the first day, the participants

were only expected to set their targets for v1 and v2 at three task rates R ∈ 0.8, 1, 1.2, both

as individuals and as a group, with v1 being a minimum of 12. This was done to discourage

all the participants from allocating all of the task on a single day, affecting the intertemporal

nature of the experiment. Different venues had different ordering for preference selection i.e.

setting individual targets first vs. group targets and vice versa to eliminate ordering effect. A

week later on day 2, participants were supposed to repeat the exercise, but were also allotted

work hours to complete v1 targets of the task after they had made their choices. They were

also informed before the start of the experiment that they had a 20% chance of receiving the

preference schedule of v1 and v2 targets set on the first day and 80% chance of decisions taken

on the second day right before the commencement of the task. They did not receive any bonus

if their specified targets of v1 and v2 were not met. Output exceeding v1 targets on the first

day was not transferable to v2. If either of the targets, v1 or v2, were not met, the bonus was

not received. They had a 33% chance of getting tasks assigned as individuals and 66% chance

of getting work assigned as a group.

To motivate the intertemporal tradeoffs faced by participants, preference decisions regarding

v1 and v2 targets had to be set for 3 different ‘task rates.’ This meant that relation between v1

and v2 varied depending on the task rate i.e. every task allocated to the later date v2 reduced

the number of tasks allocated to the sooner date v1 by a stated factor. For example, a task rate


of 1:0.8 implies that each task allocated to day 3 v2 reduces by 0.8 the number in day 2 v1. The

3 different task rates were R ∈ 0.8, 1, 1.2 and the participants were informed that they could

be assigned the proposed targets for any one task rate randomly and the chances are equally

likely. The reason for choices of values of R is that besides having a value of 1 which is the

most natural one, they also cover the situations in which a task allocated to Week 3 reduces or

increases tasks allocated to Week 2. The participant’s decision can be formulated as allocating

tasks (v1, v2) = f(R, V ) over time subject to the present-value budget constraint. The v1 and

v2 satisfied the intertemporal budget constraint

v1 +R · v2 = V.

These intertemporal bonus contracts can be used to investigate intertemporal preferences.

The allocations participants make, (v1, v2), convey information on their discounting. Additional

experimental variation permits identification of an important behavioral aspect of intertemporal

choice: the existence of present-biased preferences.

The results in this paper are important because it puts forward both nonparametric and

a parametric characterizations of individual and collective intertemporal choice for the same

set of participants, under experimentally controlled environments, based upon intertemporal

allocations of effort2. Starting with the an approach that was free of functional/structural

form using the experimentally induced exogenous variations of controlled factors, the study

moves to a theory based parametric analysis of time inconsistency issues on the individual

and group level. The reason for this approach is that our subsequent parametric estimates

are resulting from restrictive parametric assumptions rather than a failure of the underlying

theoretical framework that is free of functional form and related to an assessment of the degree

of differences between these two kinds of decision environments. In the structural part of

empirical analysis the preference structure associated with the discounted utility approach is

applied to model group behavior without modification. This is in line with representative agent

modeling structure mostly used in macroeconomics literature. This unitary approach assumes

the collective acts as a single decision making unit and therefore, can be treated as a rational

individual.

2which is in domain of consumption rather than domain of monetary choice. The monetary domain hasseveral confounding factors for the identification and estimation of the shape of time preferences. For moredetails please see Andreoni and Sprenger (2013) (forthcoming QJE) paper


3 Experimental Design

The experiment focuses on allocations of effort for one kind of job/task. This task could be

completed over time. Some portions of the task might be completed sooner, and the remaining

portion of the task could be completed later, depending on their choices and chances.

This task consisted of taking a specific number of pictures of book pages through cell phones.

As mentioned above the total was V = 200 pages for the individual setting and V = 400 pages

for the group setting. The aforementioned numbers of pictures were supposed to be taken

in one hour (9:30 pm to 10:30 pm, Pakistan Standard Time) on the second and third day of

the experiment. Participants were also instructed that their phone had a unique International

Mobile Equipment Identity (IMEI) number. Once they took a picture, they were also required

to upload the picture using the application on the phone by connecting their devices to LUMS

server’s Wi-Fi. The participants were also given free Internet data package as backup by Telenor

Pakistan Limited, which has reliable and efficient cellphone data services in the Lahore area.

Participants were instructed to ensure that their pictures were clear and that the page numbers

were legible. If the page numbers were not legible, the picture would not be counted towards

their target for the bonus purposes.

In order to avoid corner solution scenarios in the allocation decisions, when a participant or

a group decides to allocate all the tasks to day 1 or day 2, we restricted the minimum number

of pictures for the day 1 task to be 12. In the subsequent statistical analysis we took care of

this issue explicitly. Only 2.50% of the total 2196 allocation decisions v1 = 12 was observed.

On the first day of the experiment the participants were asked to make allocation decisions

such that 3 decisions (for each R) were for each individual and 3 decisions (for each R) were

in a group of two formed randomly. On the second week, they were asked again to make the

same 9 allocation decisions (3×2 individuals+3 group). Towards the end of the second session,

all the participants were informed which allocation had been selected (randomly) for them out

of 18 total decisions that they recorded over the last two days of the experiment. We called

the selected task allocation decision as “decision-that-counts”. In the second week, 2 hours

after the allocation decision session, participants were asked to complete their “decision-that-

counts” allocations in the specified timing, 9:30 pm to 10:30 pm, Pakistan Standard Time. All

the pictures taken through cellphones provided to them, were geo-stamped and time-stamped.

The allocation decisions across two weeks depended on the task rate. The task rate varied

across the decisions. On the target-setting page of the application (installed in the cell phones),

every slider bar corresponded to a specific task rate. For example in the first slider bar the task

rate was 1:0.8, such that every task the participant allocated to the second day v1 reduced the

number of tasks allocated to the third day v2 by 0.8. For simplicity, the task rates were always

represented as 1 : R, and the participants were fully informed of the value of R ∈ 0.8, 1, 1.2


when making their decisions. An image of the main page of the application is provided as figure

1.

Figure 1: Slider Bar Used to Capture Task Allocations

Notes: The above slider bars show the individual task allocation decisions over the two weeks i.e. V = 200. The blue letters intranslate (literally) “set target”. The next lines (from left to right) translates to “First day: 111; Second day: 111” for slider bar0.8, “First day: 100; Second day: 100” for slider bar 1, and “First day: 90; Second day: 90” for slider bar 1.2. The text next tothe box translates to “finalize target” and the black letters on the bar translate to “set target.”

Participants were informed about three important factors determining their “decision-that-

counts” allocated for them. First, the allocated task would be either from 13th March Decisions

(1st day of experiment) or 20th March Decisions (2nd day of experiment) according to a 20%

and 80% chance respectively. The chance of the “decision-that-counts” being from the set

of individual decisions was 33% and there was 66% chance that “decision-that-counts” would

come from their set of group allocation decisions. If the “decision-that-counts” turned out to

be a group decision, then as a group they would have to complete V = 400 pages on the 2nd

and 3rd day (27th March) of the experiment.

In the group setting, participants were also informed that there would be a group bonus

rather than individual bonuses for the members. If a participant was selected for group task

completion, her/his bonus would depend on the groups performance rather than their individual

performance within the group. We gave no instructions about the intra-group division of

workload i.e. who in that group will complete what fraction of total group task. If a group

completed the task allocated then both individuals in that group would get the bonus of $15

each, regardless of how the labor was divided.

The last factor determining their selected “decision-that-counts” was selection of three R ∈


0.8, 1, 1.2. Participants were informed that all the Rs are equally likely, either in group or

in individual settings. The randomization process, which is inline with Augenblick, Niederle,

and Sprenger (2015) ensures incentive compatibility rather than cheap talk for all decisions.

The design choice which ensured that participants had a 20% chance of receiving preference

schedule of v1 and v2 targets from the first day of decision making process was made to allow

and intimate them about their own potentially present-biased behavior.

3.1 Design Details

Our experiment consists of a single drive. This drive comprised of two sessions related to train-

ing/decision and two sessions related to task completion. For the training/decision sessions,

which began on 13th March and ended on 20th March, the physical presence of the participants

was compulsory. In these sessions, participants were taught the use of the application and the

method of sending the data to the main server. Participants were invited to eight different

venues within the LUMS campus. Each participant was randomly assigned to a particular

venue and group beforehand, and informed the night before via email to be present at that

particular venue. Questions about their demographics and big five personality traits were also

recorded during this session. Participants also took part in different laboratory games related to

prediction behavior, risk behavior, and confidence. All other activities, besides task allocation

decisions, were conducted individually i.e. participants played the games individually rather

than as a group.

3.1.1 Training and Allocation Decisions

Each session lasted 2 hours i.e. from 6 pm to 8 pm. In the last training/decision session, which

was conducted on 20th March 2015, all participants who had participated earlier in the 13th

March session were invited back. Training in all venues was identical except for one important

factor: Four out of a total of eight venues were randomly selected and in those training sessions

participants were asked to make individual task allocation decisions first and group decisions

second. Similarly, in the remaining venues, participants were instructed to make individual

task allocation decisions after group decisions. Each training session was supervised by a venue

in-charge along with one or two assistants. A hotline number was provided if assistance was

required for those facing some problems while performing tasks or during its submission due to

hardware or software issues. Towards the end of every session and before making the allocation

decision, they were given some time to practice using the cellphone application.


3.1.2 Experimental Timeline

An image of the experiment’s timeline is provided in Appendix section A4 figure 5. We started

our study on Friday March 13th, 2015 with a training/decision session. 244 participants had

been recruited and were randomized into eight different venues. Participants making the in-

dividual task allocation decisions first and the group decisions second, were assigned to four

venues; in the remaining four venues, we invited the participants who made the group decision

first and individual second. During all the training sessions, venue in-charges were instructed to

make sure that participants within the same group exchanged their email addresses and shared

their contact numbers with each other for further communication.

Sessions regarding task completion were held on Friday March 20th and Friday March 27th.

Every participant was informed that these sessions were only for one hour starting from 9:30

pm, and ending sharply at 10:30 pm. Physical presence at these training sessions was not

required. Participants could perform the task from anywhere within the LUMS campus or even

at their homes (in case of day scholars), provided that they have Wi-Fi there.

3.1.3 Effort Allocations

In Week 1 (Friday 13th March), participants allocated tasks between Week 2 (Friday 20th

March) and week 3 (Friday 27th March), individually and in groups. In Week 2, participants

again allocated tasks between Weeks 2 and 3. Participants were not reminded of their initial

Week 1 allocations in Week 2. Note that in Week 1, participants were making decisions involving

two future work dates, whereas in Week 2, participants were making decisions involving a

present and a future work date. Before making decisions in Week 1, participants were told of

the Week 2 decisions and were aware that exactly one of all Week 1 and Week 2 allocation

decisions would be implemented.

3.1.4 The Allocation-That-Counts

Each group of two randomly-paired individuals made 18 decisions allocating work to Weeks 2

and 3: 12 decisions were made individually and 6 were group decisions. 12 individual decisions

constituted from 6 decisions of two individuals in the group and 6 group decisions were joint

decisions by both members. After Week 2 decisions, one of these 18 allocations was chosen at

random as the ‘allocation-that-counts,’ and participants had to complete the allocated number

of tasks on the two work dates to ensure successful completion of the experiment.


4 Results

4.1 Sample Details

Table 1 summarizes our sample of participants regarding the important dimensions of demo-

graphics and behavior. The table shows that the mean age of the participants of this experiment

is approximately 20.3 years. In the overall sample, 39% were females. Most of the participants

had no access to formal savings accounts. In the behavioral economics empirical literature,

savings accounts have been used to predict the degrees of patience or present-bias. As men-

tioned earlier the individuals were randomly paired in a group. The mean of the group-mate

acquaintance index indicates that in most of the groups the individual members knew each

other at the start of the study. The duration of the acquaintances range from 0 – indicating

that the members have just met – up to 5 years of acquaintances. In the subsequent structural

analysis we explicitly control the time duration of acquaintance to explain group procrastina-

tion tendencies. In the context of this experiment, this variable tries to capture the effect of

coordination externality in the stage of intertemporal allocation of tasks.

The Big 5 personality index was first developed by psychologists in the 1980s and has

subsequently become the standard and most widely used personality taxonomy in the field.3

The Big 5 personality index consists of five traits, i.e. agreeableness, emotional stability (the

reverse of neuroticism), extraversion, conscientiousness, and openness.

Table 1: Summary Statistics

# of Obs Mean Standard Deviation Minimum Maximum

Demographic

Age (years) 244 20.32 1.75 17 27

Gender (Female = 0) 244 0.61 0.48 0 1

Has a On Campus Job (Yes =0) 244 0.91 0.28 0 1

Had a Savings Account (Yes =0) 244 0.73 0.44 0 1

Has a Savings Account (Yes =0) 244 0.62 0.48 0 1

Group-mate acquaintance Index (Least =1) 244 3.04 1.28 1 5

Acquaintance time duration (months) 244 13.84 21.56 0 60

Big 5

Openness 212 3.30 0.45 2.17 4.42

Conscientiousness 212 3.43 0.52 1.75 4.92

Extroversion 212 3.25 0.33 2.16 4.44

Agreeableness 212 3.44 0.46 2.33 4.58

Neuroticism 212 2.82 0.57 1.25 4.67

Notes: The Big 5 personality trait variables were recorded on a one to five Likert scale.

3See John et al. (2008) for a summary of the measures and its history. For a summary of empirical resultsin psychology and economics, see Borghans et al. (2008).


Following the methodology and the questionnaire by Callen et.al (2015), we measure these

traits using a 60 question survey developed specifically in Urdu and validated for use in Pakistan

by the National Institute of Psychology at Quaid-i-Azam University, Islamabad. For each

participant the personality traits are measured separately, using a weighted (equal) average of

12 questions for each trait. The sample stats imply that the participants had high scores in

all dimensions of personality. Although there are studies which have explicitly investigated the

relationship between these traits and time preferences, this paper has not taken that approach.

8085

9095

100105

Tasks

Allocat

ed to S

ooner D

atev1

.8 .9 1 1.1 1.2R

Panel A: Individual Decisions

160180

200220

Tasks

Allocat

ed to S

ooner D

atev1

.8 .9 1 1.1 1.2R

Panel B: Group Decisions

Immediate Choice Advance Choice

Figure 2: Discounting Behavior

Notes: Mean behavior in Individual and Group task allocated to sooner date combined.

4.2 Regression Analysis

Table 2 presents corresponding regression analysis for aggregate behavior focusing only on time

inconsistency. Given the experimental design, we regress the natural log of v1 on the dummy

variable taking the value of 1 when the allocation decision is immediate and its interaction with

individual decision, and decision order dummies. We controlled for experimentally induced

variations, i.e. natural log of R, individual decision and decision order dummies. Using the

sample of decisions received during the training sessions, we estimated the two-limit Tobit

regression. This regression specification corrects for censoring around the minimum numbers

of task (i.e. minimum 12 tasks) to avoid corner solution scenarios in the allocation decisions.

The results present a non-parametric evidence for time inconsistency. Column (1) signifies

the findings from Figure 2, individual and groups] decisions combined, demonstrating that


participants allocated significantly fewer tasks as a percentage to v1 as R increases and when

the allocation decision is immediate. In other words when the task rate is low, sooner tasks are

relatively cheap to complete, and participants allocated more tasks compared to when the task

rate is high, and sooner tasks are relatively expensive to complete. The estimate of immediate

decisions dummy is around -15% and this is statistically significant. This indicates that on

average 15% fewer tasks were chosen on the 1st day of task completion and it is statistically

signifiant (p = 3.58).

In column (2) we introduced the dummy for the individual decisions and its interaction

with immediate decision dummy variable, we observed that participants making immediate

choices in the individual setting allocate around 12% (s.e. around 4.4%) fewer tasks to v1 than

making the same decision a week before the first day of task completion. In the group setting,

the individuals allocate 21% fewer tasks on the first day of task completion compared to a

week before. Comparing the estimate of immediate decisions dummy and its interaction with

the individual decision dummy estimate implies that on the individual level participants are

more time consistent and less present biased, on average, compared to a situation in which they

decide in groups. This is clear from the estimate of immediate decision dummy when interacted

with individual decision dummy, which increased by a factor of 9%, and is significant at 5%

level. The estimate IndividualDecision(= 1) is negative and significant due to the variation in

the total number of tasks over the two weeks period between individuals and their respective

groups. In the group setting it was fixed at 400 and in the individual setting it was 200. The

negative estimate signifies the fact that as an individual the participants were to allocate fewer

tasks compared to the group setting.

These findings are contrary to recent few studies (examples include E. Carbonea, G. Infante,

2015; L. Boemont, E. Diecidue, and O. Haridon, 2015) in which authors have found or tried to

establish that groups are less present biased and more time consistent compared to individuals.

We mainly attribute this difference to three factors. First, studies in which groups are less

present biased have used money. Several confounds exist for identifying the shape of time

preferences from such monetary choices. Issues of payment reliability and risk preference suggest

that subject responses may be closely linked to their assessment of the experimenter’s reliability

rather than solely their time preferences. Also monetary payments may not be suitable in

identifying model parameters which are defined over time-dated consumption. The first two

sections of the paper by Augenblick, Niederle, and Sprenger (2014) provides a detailed discussion

on this issue.

Second in all of those studies, the experimenter usually invited couples4, where, as in our

setting, groups were formed randomly. As evident from the sample summary statistics, the

4where laboratory factors also play an important role, by virtue of structural arrangements


group acquaintance index for the pool of people, participated in our experiment averages around

3.09 with the standard deviation of 1.27. This further suggests that since group members were

not fully aware of each other’s individual characteristics, they opted for conservative targets

(to achieve the bonus amount) especially when making the immediate choice allocations. This

is also evident from the figure 2.

The last factor in which the group present bias estimates tend to be more than individual

decisions (for exactly the same people) is due to the underlying incentive structure for the bonus

achievement. Since the tasks were easy to perform and fundamentally perfectly substitutes in

nature, the effect of coordination externality could have become an important factor. In the

last part of the results section we will explore these issues in more detail by putting some

theory-based structure on the estimates.

In column (3) we analyze the effect of the order in which task allocation decisions were

taken. The ordering does not have any significant effect on the behavior in the individual

or joint decisions. The estimates of individual Decision First (=1) dummy and its interaction

with immediate decision dummy are statistically insignificant. This along with the non changing

estimates of the estimate of immediate decision and its interaction with the individual decision

dummy estimates in the last column confirms the robustness of our finding.


Table 2: Two-Limit Tobit Regression Analysis

Dependent variable: Log of Tasks Allocated to the 1st Day

(1) (2) (3) (4)

β1 : Log Task Rate -0.46*** -0.46*** -0.46*** -0.46***

(0.08) (0.08) (0.08) (0.08)

β2 : Immediate Decision (=1) -0.15*** -0.21*** -0.14* -0.20**

(0.04) (0.05) (0.07) (0.08)

β3 : Individual Decision (=1) -0.75*** -0.75***

(0.02) (0.02)

β4 : Immediate Individual Decision (=1) 0.09** 0.09**

(0.04) (0.04)

β5 : Individual Decision First (=1) -0.04 -0.04

(0.06) (0.06)

β6 : Immediate Individual Decision First (=1) -0.02 -0.02

(0.08) (0.08)

β0 : Constant 4.70*** 5.20*** 4.72*** 5.23***

(0.03) (0.03) (0.04) (0.04)

# of Obs 2196 2196 2196 2196

# of Groups 122 122 122 122

Adj R2 0.01 0.17 0.01 0.17

F-stats 23.68 326.64 13.36 219.78

H0 : β2 + β4 = 1

p-value 0.00

H0 : β2 + β6 = 1

p-value 0.00

H0 : β2 + β4 + β6 = 1

p-value 0.00

Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at group level. The table presents

the estimates of Immediate Decision (=1) and its interactions with other experimentally induced variations

using Two-Limit Tobit Regression technique. Column (1) presents aggregated decisions estimates. Column

(2) captures the estimates of Immediate Decision (=1) for the groups and individuals separately. Column

(3) represents the results of effect of decision ordering and its interaction with Immediate Decision (=1).


4.3 Structural Analysis

Motivated by our non-parametric analysis, we proceed to estimate structural time preference

parameters. Subjects allocate effort to an earlier date, v1, and a later date, v2. We assume

quasi-hyperbolic discounting. Under the assumption of a quadratic cost of effort function and

that individuals/groups discount the future quasi-hyperbolically (Laibson, 1997; O’Donoghue

and Rabin, 1999), the participant’s preferences can be written as:

b1v21 + β1d=1δk b2v

22

Normalizing b2 = 1 and therefore dividing the intertemporal effort cost function by b2 (to

get rid of the scaling effect) the above cost function can be written as:

γ v21 + β1d=1δk v2

2

Here v represents a task performed on a given day. γ > 0, and γ = 1 will imply that the

effort cost function is stationarity over time. k captures delay length, which in this experiment

was fixed at 7 days. The indicator 1d=1 captures whether the decision is made in advance or

immediately on day 1 of the task performance. The parameters β and δ summarize individual

or group discounting with β capturing the degree of present bias, active for participants who

make immediate decisions, i.e. 1d=1 = 1. If β = 1, the model nests exponential discounting

with discount factor δ, while if β < 1 the decision-maker exhibits a present bias, being less

patient in immediate relative to advance decisions.

Modeling the group decisions, we assume the group’s members are characterized by indi-

vidual preferences and that the group acts as a specific decision maker similar to an individual

whose time preference parameters could be measured independently of its members’ prefer-

ences. This modeling technique is in line with representative agent setup mostly used in macro

modeling. This unitary approach assumes that the collective acts as a single decision making

unit and therefore can be treated as a rational individual.

Minimizing discounted costs subject to the intertemporal budget constraint of the experi-

ment yields the intertemporal Euler equation:

γv∗1R = β1d=1δkv∗2. (1)

Here v∗1 and v∗2 are the optimal tasks performed on a day 1 and day 2 of the experiment. This

tangency condition implies that when individual/group preferences are dynamically consistent,

the optimal (v∗1v∗2

) does not depend on the parameter β1d=1 but only depends on the task rate

R, and the delay length k. Using the Euler equation with intertemporal budget constraint and


rearranging yields the solution function for the optimal v∗1:

v∗1 =

(β1d=1δkV

γR2 + β1d=1δk

).

and

v∗1 =

(β1d=1δkV

γR2+β1d=1δk

)d = 1

(δkV

γR2+δk

)d = 0

The above equation implies that v∗1 is a non-linear function of (R,1d=1, k, V ) 5.If we assume the

allocation decisions satisfy the above equation subject to an additive error term, ε, we arrive

at the non-linear regression equation

v1∗it = f(V,Rit,1d=1, k) + εit. (2)

Using the above equation, the parameters (β, δ, γ) can be estimated with standard tech-

niques6. The non-linear least squares (NLS) procedures (see Appendix: A1) permitting the

estimation of preference parameters at the aggregate or individual level, are implemented in

many standard econometrics packages (in our case, Stata).

Before we start with the explanation of our results, it is important to recognize the strengths

and weaknesses of such an NLS preference estimator. Parameter γ can be estimated as opposed

to assumed, which is an advantage. A potential disadvantage is that the NLS estimator is not

well-suited to the censored data issues inherent to potential corner solutions without additional

assumptions. In the context of this study, as mentioned above, the corner solution scenario

arose only in a few cases. For 2.50% of the total 2196 allocation decisions, v1 = 12 was

observed. Although the NLS estimator can be adapted to account for possible corner solutions

by adapting the criterion function and making additional distributional assumptions, in this

paper we did not take this approach. Andreoni and Sprenger (2012) suggest a complementary

approach: taking logs of the Euler Equation allows one to estimate the model parameters in

a two-limit Tobit framework that corrects for censoring. As expected, given the total number

of corner points in the data, the results in Table 7 (Appendix: A1), which are based on the

method by Andreoni and Sprenger (2012), are nearly identical to the results in Table 3.

Table 3 columns present non-linear least squares regressions’ estimates with standard errors

5For this class of effort cost function, both relative risk aversion and intertemporal elasticity of substitutionare function of v.

6please see the appendix for more detail


Table 3: Non-Linear Least Squares Analysis

Dep. Var: v1∗it

Combined Ind V s. Group Decision Order

βc 0.78*** βI 0.82*** βIF 0.77***(0.05) (0.06) (0.05)

δc 0.98*** δI 0.96*** δIF 0.98***(0.04) (0.03) (0.06)

γc 1.07*** γI 1.00*** γIF

1.17***(0.23) (0.22) (0.39)

βG 0.71*** βIS 0.0.79***(0.06) (0.10)

δG 1.01*** δIS 0.95***(0.04) (0.03)

γG 1.23*** γIS 0.85***(0.29) (0.19)

# Observations 2196 # Observations 2196 # Observations 2196# Groups 122 # Groups 122 # Groups 122RMSE 0.54 RMSE 0.54 RMSE 0.54LL -1781 LL -1199 LL -578H0 : βc = 1χ2(1) = 16.01,p-value=0.000

H0 : βI = 1χ2(1) = 9.240,p-value=0.002

H0 : βIF = 1χ2(1) = 19.10,p-value=0.000

H0 : δc = 1χ2(1) = 0.500,p-value=0.481

H0 : βG = 1χ2(1) = 23.75,p-value=0.000

H0 : βIS = 1χ2(1) = 4.920,p-value=0.026

H0 : γc = 1χ2(1) = 0.100,p-value=0.757

H0 : βI = βGχ2(1) = 4.530,p-value=0.033

H0 : βIF = βISχ2(1) = 0.010,p-value=0.907

Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors clustered at the group level. χ2(1)0.05 = 3.84.The table presents the structural estimates of intertemporal hyperbolic discounting model using Non-linear LeastSquares Regression technique. Column (1) presents the combine decisions estimates of β, δ, and γ. Column (2)captures the structural estimates of the model for the groups and individuals separately. Column (3) represents thestructural estimates based on the effect of decision ordering.


clustered at the group level. In column (1), we find the estimated cost parameters’ ratio

γ = 1.07(0.23). The null hypothesis of stationary cost of effort function could not be rejected,

χ2(1) = 0.10 < 3.84. The estimated weekly discount factor is averaged around 0.98 and under

the null hypothesis that H0 : δ = 1 χ2(1) = 0.50 < 3.84 implies that weekly discount factor

of 1 can not be rejected at 5% level of significance. Regarding our main parameter of interest

β, we estimate an aggregate β = 0.78(0.05), and reject the null hypothesis of dynamic/time

consistency, H0 : β = 1 χ2(1) = 16.01 > 3.84. In column (1), our estimate of time consistency

β is closer to the results of table 2, column (1) of non-parametric results.

In column (2) of the table the time preference parameters are estimated separately for

individuals and groups, using the full sample of decisions. One can observe that the estimate

of the individuals’ present bias parameter is less than the groups’ estimate. Both estimates are

statistically significant at 5% level of significance. Under the null hypothesis that H0 : βI = 1,

χ2(1) has a value of 9.24>3.84 and H0 : βG = 1, χ2(1) has a value of 23.75>3.84. χ2(1) values

for both hypotheses imply that individuals and groups allocation decisions are present biased.

Similarly, the estimate of individuals’ weekly discount parameter is less than groups’ parameter

estimate. Both estimates are statistically significant at 5% level of significance. Under the

null hypothesis of dynamic/time consistency, H0 : δI = 1, χ2(1) has a value of 1.40<3.84 and

H0 : δG = 1, χ2(1) has a value of 0.05<3.84 implying that individual and group allocation

decisions can have a weekly discount factor of 1. This finding is in line with the studies’

estimates of discount factor in the task domain.

Testing the individual estimate of present biased estimate against the group estimate, the

null hypothesis test of H0 : βI = βG χ2(1) = 4.53 is rejected. Using Stata based lincom

command, which computes point estimates, standard errors, t-statistics, and p-values for a

specified linear combination of coefficient estimates, βI − βG has a coefficient of 0.11 with the

p-value of 0.03. On the other hand H0 : δI = δG χ2(1) = 5.12 is rejected and βI − βG has

a coefficient of -0.05 with the p-value of 0.02. These results are similar to our non-parametric

analysis where the degree of dynamic/time inconsistency is, on average, more in group decisions

compared to individual decisions. In other words, this result implies that group task allocation

decisions are more present biased compared to individual settings.

Regarding discount factor estimates and their differences, the results are consistent with

Milch et al. (2009) finding that participants discount more as individual decision makers than

in a group decision context. This finding that, the patient decisions taken by groups also show

that the discount factors for groups are more and in line with market interest rates than the

discount factor for individuals.

Comparing the cost parameter estimates’ between the individual and group settings we

observe that while moving from individual to the group setting, the estimate of γ increases but


γI − γG has a coefficient of -0.23 with the p-value of 0.13 which indicates that the difference

is not statistically significant. The individual decisions’ estimated cost parameters’ ratio γI =

1.00(0.21). The null hypothesis of stationary cost of effort function could not be rejected,

since χ2(1) test has a p-value of 0.99. Finally for the groups’ estimated cost parameters’ ratio

γG = 1.23(0.29). Under the null hypothesis of groups’ stationary cost of effort function also

could not be rejected, since χ2(1) test has a p-value of 0.43. These results indicate that the

underlying cost of effort functions for both individuals and groups decisions are stationary.

In column (3) of the table, the time preference parameters are estimated for ordering effect

using the full sample of decisions. One can observe that the estimate of present bias in decision

setting, where individual task allocation decisions preceded the group task allocation decisions,

present bias estimate in the former is lower than in the later, but the difference is statistically

insignificant. The discount factor in the former is higher than the later but again the difference

is insignificant statistically. These results also confirm the results in table 2 of non-structural

estimates.

Two questions naturally arise in the current context, as discussed in the introduction part,

although the hyperbolic discount factor is microfounded in the misalignment of preferences

between two individual members. First, what are the relevant structural channels through

which one can pin down and connect the more pronounced (compared to individual) group

procrastination tendencies to its group members? Second, are there other non-theory based

important variables through which one can explain the group procrastination tendency beyond

the structural channels? The answers to these questions are important for understanding

the empirical validity of the aforementioned theories and other channels which can broaden

our knowledge of group based decisions and their dynamics. In the next section we start

discussing the theoretical background connecting the groups decision process to their individual

constituents.

4.4 Theoretical Background: Individual vs. Group

As mentioned in the introduction part of the paper that heterogeneity across members in a group

context is natural assumption for many applications. It is relevant for households, and prac-

tically any group/committee of agents making intertemporal decisions, including legislature,

management teams, and in the context of this experiment the task/effort allocations. Given

the sets of results obtained in the non-parametric and structural part (specifically groups are

more present biased compared to their corresponding individual members) of the main results

section, here in this part of the paper we would like to introduce some theoretical background

for the group decisions. We will try to connect the individual decisions with the group alloca-

tion decisions observed in this experiment and using these theory based connections we test a


set of hypothesis mentioned in the introduction part of the paper.

4.4.1 Collective Decision Functions

A collective cost of effort function can be thought of as providing a “planner’s” cost of effort

function for a group. Examples include taking a weighted average of the agents’ effort functions

(F [C](v) =∑

i ωiCi(v)) where C is cost function and v is effort provision. This is an example of

utilitarian approach; other examples would include minimum of agents’ cost of effort provision

(F [C](v) = miniCi(v)) which would be a Rawlsian approach. In the context of this experiment

we know C = (δ1, β1, γ1; ..; δn, βn, γn). Here we consider an important class of collective cost of

effort functions: those that time− separable. Jackson and Yariv (2015) show that this class of

functions exhibit a particular sort of time inconsistency or intransitivity7: present bias, which

matches the empirical evidence presented in the paper.

As postulated by Jackson and Yariv (2015) given all the participants in the experiment are

time consistent for any profile (δ1, γ1; ..; δn, γn) ∈ Cn, a time-separable collective/group cost of

effort function takes the form

F [δ1, γ1, ; ...δn, γn](v) =∑t

δtC(vt).

such that δt =∑

i ωiδti . Time-separable group/collective functions are often used in the liter-

ature. For instance, standard utilitarian aggregation of individual utilities (or one that puts

different weights on different individuals) is a special case. According to Jackson and Yariv

(2015) for any profile (δ1, γ1; ..; δn, γn) ∈ Cn such that for some k, j, δk 6= δk, a collective

function of the form

F [δ1, γ1, ; ...δn, γn](v) =∑t

δtC(vt).

such that δt =∑

i ωiδti for each t, and so

F [δ1, γ1, ; ...δn, γn](v) =∑i

ωiCi(v).

F is either dictatorial or present biased8. It is pertinent to mention that before taking the

group decisions every participant was asked to make unanimous decisions in group setting

since the share of bonus was fixed for each group member, therefore the collective discount

7The intransitivity here is quite different from Condorcet’s (1785) description of the voting paradox andArrow’s impossibility Theorem (1963) because our collective decision structure are quite different from thevoting setting mentioned in these papers.

8for the detail of proof please see Jackson and Yariv (2012) and Jackson and Yariv (2015) papers


factor must be a weighted sum of the participants, and so must correspond to a weighted

utilitarian collective/group cost of effort function.

The proposition encompasses many of the formulations of time inconsistent preferences. In

the structural analysis part of the paper we assumed quasi-hyperbolic formulation which in

this case corresponds to δ1 = 1 and δt = βδt−1 for all t > 1. As long as behavior has separable

structure and satisfies unanimity, the proposition shows that a present bias is to be expected.

Using this proposition, a set of testable hypotheses can be generated and tested subse-

quently. In this section using the groups’ time preference estimates (specifically focusing on

groups’ procrastination tendency related to present bias estimate) and their individual mem-

bers’ structural estimates (including discount rate, present bias, and effort cost parameters’

estimates), the roles and effects of theory based heterogeneities are explored empirically on

group procrastination tendency. These within group heterogeneities include the differences in

individual members’ (i) discount factor, (ii) parameter governing the cost of effort, and (iii)

bargaining position of each person in the group.

Regarding bargaining powers we know that in a utilitarian formulation, group decisions

depend on the preferences of the individual members, and the relative strengths of the individual

members’ weights in the group decisions captured by Pareto/bargaining weights. Empirically

the introduction of a bargaining mechanism into the group decision making process by Manser

and Brown (1980) and McElroy and Horney (1981), there has been a development of so-called

collective models, which assume that groups can achieve efficient decisions (Chiappori, 1992;

Browning and Chiappori, 1998). According to Browning and Chiappori (1998), in the context

of this experiment the group’s intertemporal effort cost function can be expressed as:

CG = ωiCi + ωjCj.

where

ωi + ωj = 1. and (ωi, ωj) > 0

These restrictions satisfy the unanimity condition in the group/collective decision making pro-

cess.

Cl = γl v21l + βl

1d=1δlk v2

2l and l =G, i, j

where CG is the group group’s intertemporal effort cost function, Ci and Cj represent the

individual members i and j intertemporal effort cost function respectively, and ωi and ωj denote

members i and j bargaining power respectively, which measures how individual preferences are

aggregated into groups’ joint decisions. In our experimental setting since we observe both

individual and joint decisions, it means that we can measure/estimate to what extent each


member influences the joint/group decisions. Using a multinomial logit model, also known

as polytomous logistic regression ωi and ωj are estimated by running the following regression

equation, for each group and summary statistics are presented in Table 6.

v1Gp = ωiv1ip + ωjv1jp + εp.

s.t. ωi + ωj = 1. and p =

1, 2, 3, ..., 122

After estimating (ωi, ωj) for each group we contructed |∆(ωIND

)| which is the absolute

difference between the groups’ individual members’ bargaining/Pareto weights. Using these

absolute difference between the individual members we further constructed (|∆ω| ≈ 1)G(= 1)

which is the dummy indicator for the group in which one of the member has a approximately

all the bargaining power. The summary statistics of these dummy indicators are presented

in Table 6. The roles and effects of these variables on group procrastination tendency are

described in Table 7.

4.4.2 Individual vs. Group Analysis

Table 4 presents within group estimates of discounting, present bias, and effort cost parameters

at the individual and group level. For each group and its corresponding individuals, we estimate

the parameters of equation (2). These parameters are estimated by nonlinear least squares as

in Table 3. Time preferences and effort cost parameters are estimable for 122 groups and their

244 individual participants. The results are broadly consistent with those estimated at the

aggregate level. The median estimated weekly discount rate is 0.92, close to the aggregate

values obtained in Table 3. In line with the aggregate results, the median individual present

bias estimate is less than the median estimated group present bias parameter. The median

estimated cost of effort parameter is 0.69, suggesting that individual decisions’ parameter is

very similar to group parameter. In addition to median values, Table 4 reports the fifth to

ninety-fifth percentile range for group estimates, and its corresponding individuals members’

estimates of the annual discount rate δ, β, and γ along with the minimum and maximum values.


Table 4: Discounting, Present Bias, and Effort Cost Parameter Estimates

N Median 5th Percentile 95th Percentile Minimum Maximum

Group

δ 122 0.92 0.75 0.99 0.55 0.99

β 122 0.90 0.18 2.32 0.02 10.8

γ 122 0.69 0.19 1.33 0.14 2.80

Individual

δ 244 0.92 0.76 0.97 0.66 1.00

β 244 0.97 0.13 2.79 0.02 15.1

γ 244 0.69 0.18 1.51 0.06 2.80

Notes: NLS estimators as in Table 3.

For the majority of individuals and groups, the employed estimation strategy generates

reasonable parameter estimates. However, extreme observations do exist. Figure 3 presents

histograms of time preference, β and discounting estimates δ. The histograms demonstrate that

a large proportion of subjects have low discount rates and maximum present bias. Estimation

results for all subjects are presented in Appendix Tables A2.

05

1015

20

0 1 2 3 0 1 2 3

Group Individual

Percen

t

Estimated present bias

010

2030

.6 .8 1 .6 .8 1

Group Individual

Percen

t

Estimated weekly discount factor

Histogram of estimated time preference parameters

Figure 3: Estimates Histogram

For further systematic analysis presented in the subsequent part, and since each group

consisted of two individuals paired randomly, every group that can be considered estimates in

terms of minimum, and maximum. Table 5 presents across groups’ estimates of minimum and

maximum discounting, present bias, and effort cost parameters, focusing only on individual

decisions. Comparing the results in Table 5 with group parameter estimates presented in Table

4, it is clear that the group median parameter estimates in Table 4 is always greater than the


minimum and less than the maximum of the median of these estimates presented in Table 5.

The same observation is also true for the 5th Percentile and 95th Percentile estimates of groups

compared with their minimum and maximum counterparts. In other words the median, 5th,

and 95th Percentile group estimates are in between the minimum and maximum of individuals

estimates. This finding is inline with Gollier and Zechhausers (2005) model of aggregation of

time preferences, in which the rate of impatience of the representative agent is a weighted mean

of individual rates of impatience, although at the extreme points this may not hold.9

Table 5: Summary of With-in Group Min and Max Parameter Estimates

N Median 5th Percentile 95th Percentile Minimum Maximum

Minimum

δ 122 0.88 0.72 0.95 0.66 0.95

β 122 0.81 0.09 1.23 0.03 11.5

γ 122 0.49 0.15 1.12 0.57 2.80

Maximum

δ 122 0.95 0.79 0.99 0.75 1.00

β 122 1.07 0.39 3.93 0.03 15.1

γ 122 0.79 0.28 1.74 0.14 2.80

Notes: Based on NLS estimators as in Table 3.

Figure 4 shows that the distribution of time preference parameters i.e. present bias and

weekly discount factor for the minimum and maximum in the group, along with its correspond-

ing collective estimates. The figure highlights two important results visually. First groups’

time preference estimates are bounded by their individual members’ estimates and secondly,

the group present bias estimates tend to be closer to within group minimum estimate. In other

words the person who has more procrastination tendency in the group dominates the overall

group procrastination tendency.

4.4.3 Relationship Between Individual and Group Parameter Estimates

In order to investigate the reasons behind/as to why groups’ task allocation decisions are

more present biased compared to their individual counterparts, we carried out a theory-based

systematic inquiry. In this last part of our empirical analysis, we put different theories to the

test. As mentioned earlier theoretically, in a group context, inconsistencies can arise simply

from the aggregation of heterogeneous preferences. Variations in individual discount rates and

cost functions’ parameters represent relevant considerations in this regard. In this part, using

9The authors also showed that heterogeneous individual exponential discounting yields collective hyperbolicdiscounting


.2.4

.6.8

1(Pro

b <= V

alue)

0 5 10 15Estimated present bias

0.2

.4.6

.81

.5 .6 .7 .8 .9 1Estimated weekly discount factor

Cumulative Distribution Function

Group Within group maxWithin group min

Figure 4: Estimated CDF

the NLS based individual and group estimates of time preference parameters’ we test Jackson

and Yariv (2015)’s main idea.

In this section we explicitly test how variations in individual discount rates, effort cost pa-

rameters, and differences in (within) groups’ bargaining weights are related and could generate

aggregate behavior which is more present biased. We also test other important individual fac-

tors shape the group behavior along with their interconnection. As discussed earlier, this paper

also explores the role of coordination externality and its effect on the group procrastination

tendencies. The mean of the group-mate acquaintance index mentioned in Table 1 indicates

that in most of the groups, the individual members knew each other at the start of the study.

The duration of the acquaintance ranges from 0 (indicating that the members have just met) up

to 5 years of acquaintance. In this section, we explicitly control the time duration of acquain-

tance to explain group procrastination tendencies, in order to see the effect of coordination

externality.

Table 6 summarizes the NLS based estimates in which we are interested and tries to establish

and test the theory-based relationship to group procrastination tendency. |∆(δIND

)| is the

absolute difference between the groups’ individual members’ weekly discount rate and it has a

mean value of 6% with the standard deviation of 0.07. The variable |∆(γIND

)| is the absolute

difference between the groups’ individual members’ cost of effort parameters and it has a mean

of 0.36 with standard deviation equal to 0.41. |∆(ωIND

)| is the absolute difference between

the groups’ individual members’ bargaining/Pareto weights. This variable has mean of 0.85

indicating that within groups the members do have different bargaining power and for most of

the groups the chances of having a non-dictatorial setup is quite high.


Table 6: Summary Statistics

Variables Mean Standard Deviation Minimum Maximum

|∆(δIND

)| 0.06 0.07 0 0.26

|∆(γIND

)| 0.36 0.41 0 2.15

|∆(ωIND

)| 0.85 0.29 0 1

|(∆ω ≈ 1)|G(= 1) 0.72 0.44 0 1

(β ≈ 1)both(= 1) 0.05 0.21 0 1

Acquaintance Time Duration (months) 13.84 21.56 0 60

Notes: No of observations is 244. |∆(δIND

)| is the absolute difference between the groups’ individual members’

weekly discount rate. |∆(γIND

)| is the absolute difference between the groups’ individual members’ cost of effort

parameters. |∆(ωIND

)| is the absolute difference between the groups’ individual members’ bargaining/Pareto

weights. (|∆ω| ≈ 1)G(= 1) is the dummy indicator for the group in which one of the member has a approximately

all the bargaining power. (β ≈ 1)both(= 1) is the dummy indicator which takes the value of 1 for those groups in

which both members are time consistent approximately.

Using ω we can construct a dummy indicator (|∆ω| ≈ 1)G(= 1) for the group in which

one of the member has approximately all the bargaining power (with |∆(ωIND

)| > 0.98). This

variable has mean of 0.72 signifying the fact that in majority of the groups the chances of

having a dictatorial member (borrowing the terminology from Jackson and Yariv; 2015) is high

i.e. they ignore the preferences of all but one agent.

Lastly it is pertinent to mention that, in the context of our design, the possibility that

in groups both participants are individually time consistent does not poses any additional

challenge for our main experimental findings since we observe both group and individual deci-

sions. Focusing on how aggregation relates to time inconsistency, we explicitly controlled for

the underlying individual preferences to isolate the effects of aggregation. In the case where

both participants are individually time consistent we constructed (β ≈ 1)both(= 1) which is

the dummy indicator taking the value of 1 for those groups in which both members are time

consistent approximately. It has a mean of 0.05 indicating that in the overall sample 5% of

groups have both members who are nearly time consistent i.e. both members have β between

0.95 and 1.05.

Table 7 presents the results of theory based inter-connections. The estimates in column

(1), and (4) of Table 7 signifies the fact that weekly discount factor heterogeneities of the

individual do explain group procrastination tendencies. These columns also show that the

presence of a dictator or dominant individual based on his/her bargaining power improves the

procrastination tendency in group. These results are inline or a direct test of Jackson and

Yariv (2015) in which they showed that for a uniform distribution of discount rates in an

otherwise homogeneous population, group utility maximization generates aggregate behavior

that corresponds to hyperbolic discounting and if there is some fundamental heterogeneity in

temporal preferences by way of differing discount factors, then the only well-behaved collective


utility functions that are both time consistent and respect unanimity are dictatorial. For these

columns under H0 : Constant = 1, F (1, 121) have p− values of 0.87 and 0.65 respectively, we

see that, given there are no variations in individual discount rates, and in effort cost parameters

the group’s allocation decisions would represent time consistent pattern given the individuals

are exponential discounters.

Column (2), and (5) of the results signifies the effect of coordination externality on group

present bias estimate. As the Acquaintance Duration between the individual members in-

creases, it is natural to think that the individual’s coordination problems lessens. This in

turn decreases the group present bias tendency. In both columns, its estimate is marginally

significant at 10%. Under the H0 : Constant = 1, F (1, 121) has a p − values shows that,

given that there are no variations in individual discount rates and in effort cost parameters,

the group’s allocation decisions would represent time consistent patterns even if the individuals

are exponential discounters with no Acquaintance Duration. Lastly, in column (3), and (6)

the association of Big5 personality traits with group procrastination tendency are investigated.

These columns show that Big5 personality traits do not have the power to explain the groups’

tendency for procrastination. The effect of presence of within group dictator is not significant

in these columns and this may be attributable to fluctuation of overall sample size.


Table 7: Individual (IND) vs. Group Regression Analysis

Dependent variable: βG

(1) (2) (3) (4) (5) (6)

|∆(δIND

)| -3.84** -3.07** -3.16** -3.86** -3.00** -3.09**

(1.25) (1.19) (1.28) (1.26) (1.23) (1.32)

|∆(γIND

)| 0.14 0.21 0.18 0.13 0.24 0.19

(0.26) (0.24) (0.25) (0.28) (0.24) (0.25)

|∆(ωIND

)| -0.25 -0.17 0.07 -0.21 -0.21 0.06

(0.23) (0.27) (0.31) (0.27) (0.26) (0.28)

(|∆ω| ≈ 1)G(= 1) 0.58** 0.43** 0.30 0.57** 0.43** 0.30

(0.20) (0.19) (0.20) (0.21) (0.18) (0.18)

(β ≈ 1)both(= 1) -0.27 -0.36 -0.38 -0.30 -0.30 -0.35

(0.23) (0.30) (0.32) (0.30) (0.30) (0.32)

Acquaintance Duration 0.01* 0.01* 0.01* 0.01*

(0.01) (0.01) (0.01) (0.01)

|∆(Big 5)| -0.08 -0.05

(0.14) (0.15)

|∆(Age (in years))| -0.04 0.02 0.06

(0.06) (0.05) (0.07)

|∆(Gender)| -0.09 0.17 0.11

(0.22) (0.18) (0.20)

Constant 1.02*** 0.76*** 0.72** 1.11*** 0.65** 0.56

(0.14) (0.19) (0.25) (0.24) (0.24) (0.34)

# of Groups 122 122 109 122 122 109

Adj R2 0.06 0.30 0.30 0.07 0.31 0.31

RMSE 1.13 0.98 1.02 1.13 0.97 1.02

H0 : Constant = 1

p-value 0.87 0.19 0.27 0.65 0.15 0.19

Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at group level. Column

(1) presents the estimates of variations in group members’ weekly discount factors, cost of effort

parameters, bargaining power estimates, presence of dictator, and both time consistent members

dummy . Column (2) presents the estimates of effect of acquaintance duration on group procras-

tination tendency controlling for the variables in column (1). Column (3) captures the estimates

of effect of within group differences in Big5 personality traits controlling for the variables in col-

umn (2). Column (4), (5), and (6) represent the estimates of theory and non-theory based factors

controlling for within group differences in age and gender.

To summarize, the main message of Table 7 is that groups whose members have misaligned

discount rates will be present biased, and the presence of a dictator within group improves


the group present biased estimate. To emphasize that divergent preferences within the group

are sufficient to render time inconsistency, even controlling for time consistent individuals and

presence of a dictator within the group, the variation in discount factors has significantly

affected group procrastination tendency. Similarly, the effect of coordination externality, which

is captured by Acquaintance Duration variable, is also important in understanding the group

procrastination tendency.

5 Conclusion

This paper proposed an analysis of individual and collective decisions through the preference

elicitation method over unpleasant task consumption. The study uses experimental data to

analyze task consumption decisions by groups of individuals who have to reach a consensus

regarding allocation of tasks over time. For this purpose a joint experimental elicitation of time

preferences was performed for the groups as well as for their individual members.

The main results of the paper are the following. First ,on aggregate, a present bias exists

in participants behavior i.e. the participants’ intertemporal allocation decisions exhibited time

inconsistency. Participants on day 2 of the experiment allocated significantly fewer tasks to v1

than on day 1 suggesting that the decisions are present biased. Second, the degree of present

bias was more pronounced in a group’s task allocation decisions as compared to an individuals

task allocation setting. This justifies the rationale for not forming the group for more or less

perfectly substitutable tasks. Third, the order in which decisions were made, whether making

the individual task allocation first and then the group task allocation or vice versa had no

effect on the degree of present bias. Lastly, using within-groups estimates of present bias and

discount factor, the variations in group’s individual members discount rates do explain group

procrastination tendencies as postulated by Jackson and Yariv (2015).

This paper acknowledges that the results could be partly explained, by a selection bias. In

our experiment, as in any experiment involving longitudinal measures, subjects were supposed

to commit to three sessions over a time span of three weeks. Here, a specificity of our subjects

is probably their ability to commit and schedule (Frederick, 2005; Perez-Arce, 2011; Dohmen

et al., 2010). The estimates of present bias and discount rates for individual choices we found

are no higher than that found in the literature, although the empirical literature on task

consumption is very limited. Moreover, we were mainly interested in comparisons. It is plausible

that the selection bias impacted all decisions to a similar extent, thus we have no big effect on

our comparisons. Finally, our coordinating device allowed groups to quickly converge towards a

given decision. In this respect, our results have implications of the way boards and committees

can achieve consistent decisions.


References

Andersen, Steffen, Glenn W. Harrison, Morten I. Lau, and Elisabet E. Rutstrom (2008):

“Eliciting Risk and Time Preferences,” Econometrica, 76 (3), 583-618.

Andreoni, James and Charles Sprenger (2012): “Estimating Time Preferences with Convex

Budgets,” American Economic Review, 102 (7), 3333-3356.

Andreoni, James and Charles Sprenger (2012): “Risk Preferences Are Not Time Preferences,”

American Economic Review, 102 (7), 3357-3376.

Arrow, K. J. (1963): “Social Choice and Individual Values,” Second Edition, Cowles Founda-

tion for Research in Economics: Yale University.

Ashraf, Nava, Dean Karlan, and Wesley Yin (2006): “Tying Odysseus to the Mast: Evidence

from a Commitment Savings Product in the Philippines,” Quarterly Journal of Economics,

121 (1), 635-672.

Augenblick, Ned, Muriel Niederle, and Charles Sprenger (2015):“Working Over Time: Dy-

namic Inconsistency in Real Effort Tasks,” Quarterly Journal of Economics.

Benjamin, Daniel, James Choi, and Joshua Strickland (2010): “Social Identity and Prefer-

ences,” American Economic Review, 100 (4), 1913-1928.

Bauer, Michael, Julie Chytilova, and Jonathan Morduch (2012): “Behavioral Foundations

of Microcredit: Experimental and Survey Evidence from Rural India,” American Economic

Review, 102 (2), 1118-1139.

Benzion, U., A. Rapoport, and J. Yagil (1989): “Discount rates inferred from decisions: An

experimental study,” Management science, 35, 270-284.

Bleichrodt, H. and M. Johannesson (2001): “Time preference for health: A test of stationarity

versus decreasing timing aversion,” Journal of Mathematical Psychology, 45, 265-282.

Borghans, Lex, Bart H.H. Golsteyn and James J. Heckman (2008): “IQ, Achievement and

Personality. Unpublished manuscript, University Maastricht

Browning, M. and P. Chiappori. (1998): “Efficient Intra-household Allocations: A General

Characterization and Empirical Tests.” Econometrica, 66(6): 12411278.

Cagetti, Marco (2003): “Wealth Accumulation Over the Life Cycle and Precautionary Sav-

ings,”Journal of Business and Economic Statistics, 21 (3), 339-353.


Callen, Michael., Gulzar, Saad., Hasanain, Ali., Khan, Yasir., Rezaee, Arman., (2015): “Per-

sonalities and Public Sector Performance: Evidence from a Health Experiment in Pakistan,”

NBER Working paper No. 21180.

Carbone, Enrica., Infante, Gerardo,. (2015): “Are Groups Better Planners Than Individuals?

An experimental Analysis,” Journal of Behavioral and Experimental Economics, vol. 57, p.112-

119.

Carlsson, F., H. He, P. Martinsson, P. Qin, and M. Sutter (2012): “Household decision making

in rural China: Using experiments to estimate the influences of spouses,” Journal of Economic

Behavior Organization.

Chiappori, P. (1992): “Collective Labor Supply and Welfare.” Journal of Political Economy,

100 (3): 43767.

Coller, M. and M. B. Williams (1999): “Eliciting individual discount rates,” Experimental

Economics, 2, 107-127.

Condorcet, Marquis de. (1785): “Essai sur l’application de l’analyse a la probabilite des de-

cisions rendues a la probabilite des voix, Paris: De l’imprimerie royale, translated in 1976 to

“Essay on the Application of Mathematics to the Theory of Decision-Making.” in Condorcet:

Selected Writings, Indianapolis, IN: Bobbs-Merrill. K. M. Baker, editor.

Cubitt, Robin P. and Daniel Read (2007): “Can Intertemporal Choice Experiments Elicit

Preferences for Consumption,” Experimental Economics, 10 (4), 369-389.

DellaVigna, S. (2009): “Psychology and Economics: Evidence from the Field,” Journal of

Economic Literature, 47, 315-372.

Denant-Boemont, L. and Y. Loheac (2011): “Time and Teams: An experimental Study about

Group Inter-temporal Choice,” Working Paper.

Denant-Bomont, Laurent and Diecidue, Enrico and l’Haridon, Olivier (2013): “Patience and

Time Consistency in Collective Decisions,” INSEAD Working Paper No. 2013/109/DSC.

Dohmen, T., A. Falk, D. Huffman, and U. Sunde (2010): “Are Risk Aversion and Impatience

Related to Cognitive Ability?” American Economic Review, 100, 1238-1260.

Dupas, Pascaline and Jonathan Robinson (2013): “Why Don’t the Poor Save More? Evidence

from Health Savings Experiments,” American Economic Review, 103 (4), 1138-1171.


Fishburn, P. C. and A. Rubinstein (1982): “Time preference,” International Economic Review,

23, 677-694.

Frederick, S., G. Loewenstein, and T. O’donoghue (2002): “Time discounting and time pref-

erence: A critical review,” Journal of economic literature, 40, 351-401.

Gine, X., J. Goldberg, D. Silverman, and D. Yang (2014): “Revising Commitments: Field

Evidence on the Adjustment of Prior Choices,” working paper, World Bank.

Gollier, C. and R. Zeckhauser (2005): “Aggregation of heterogeneous time preferences,” Jour-

nal of Political Economy, 113, 878-896.

Halevy, Y. (2015): “Time Consistency: Stationarity and Time Invariance,” Econometrica, 83,

335-352.

Harrison, Glenn W., Morten I. Lau, and Melonie B. Williams (2002): “Estimating individual

discount rates in Denmark: A field experiment,” American Economic Review, 92 (5), 1606-

1617.

Hausman, Jerry A. (1979): “Individual Discount Rates and the Purchase and Utilization of

Energy Using Durables,” The Bell Journal of Economics, 10 (1).

Holcomb, J. H. and P. S. Nelson (1992): “Another experimental look at individual time

preference,” Rationality and Society, 4, 199-220.

Jackson, M. and L. Yariv (2014): “Present Bias and Collective Dynamic Choice in the Lab,”

American Economic Review, 104, 4184-4204.

Jackson, M. and L. Yariv (2015): “Collective Dynamic Choice: The Necessity of Time Incon-

sistency,” American Economic Journal: Microeconomics, 7.

Janet Metcalfe and Walter Mischel (1999): “A Hot/Cool System Analysis of Delay of Grati-

fication: Dynamics of Willpower,” Psychological Review, 106, p3-19.

John, O. P., Naumann, L. P., Soto, C. J. (2008): “Paradigm shift to the integrative Big Five

trait taxonomy: History, measurement, and conceptual issues.” In O. P. John, R. W. Robins,

L. A. Pervin (Eds.), Handbook of personality: Theory and research (3rd ed., pp. 114158).

New York, NY: Guilford.

Kang, M.-I. and S. Ikeda (2014): “Time discounting and smoking behavior: evidence from a

panel survey,” Health Economics, 23, 1443-1464.


Kerry Milch, Elke Weber, Kirstin Appelt, Michel Handgraaf, and David Krantz (2009): “From

individual preference construction to group decisions: Framing effects and group processes.”

Organizational Behavior and Human Decision Processes, 108(2): 242255, 2009.

Kirby, K. N. and N. N. Marakovic (1995): “Modeling myopic decisions: Evidence for hyper-

bolic delay discounting within subjects and amounts,” Organizational Behavior and Human

Decision Processes, 64, 22-30.

Koopmans, Tjalling C. (1960): “Stationary Ordinal Utility and Impatience,” Econometrica,

28 (2), 287-309.

Kuhn, Michael A (2013): “Curing the Calorie Crunch: The Effect of EBT on Household

Present-Bias,”

Laibson, David (1997): “Golden Eggs and Hyperbolic Discounting,” Quarterly Journal of

Economics, 112 (2), 443-477.

Laibson, David, Andrea Repetto, and Jeremy Tobacman (2005): “Estimating discount func-

tions with consumption choices over the lifecycle,” Working Paper.

Lawrance, Emily C. (1991): “Poverty and the Rate of Time Preference: Evidence from Panel

Data,” Journal of Political Economy, 99 (1), 54-77.

Manser, M. E., and M. Brown. (1980): “Marriage and Household Decision-Making: A Bar-

gaining Analysis.” International Economic Review, 21(1): 31-44.

Martin S. Feldstein (1964): “The Social Time Preference Discount Rate in Cost Benefit Anal-

ysis, The Economic Journal, 74, p360-379.

Mazzocco, Maurizio (2008): “Intertemporal Behavior and Household Structure,” Manuscript,

Department of Economics, Duke University.

Mazzocco, Maurizio (July 2007): “Household Intertemporal Behavior: a Collective Character-

ization and a Test of Commitment,” Review of Economic Studies, Vol. 74, No. 3, pp. 857-895.

McElroy, M. B. and M. J. Horney. (1981): “Nash-Bargained Household Decisions: Toward a

Generalization of the Theory of Demand.” International Economic Review, 22(2): 333-49.

O’Donoghue, Ted and Matthew Rabin (1999): “Doing it Now or Later,” American Economic

Review, 89 (1), 103-124.

Perez-Arce, F. (2011): “The Effect of Education on Time Preferences,” Working paper, RAND

Corporation.


Richard Thaler and Hersh Shefrin (1981): “An Economic Theory of Self Control,” Journal of

Political Economy, 89.2, p392-410.

Samuelson, Paul A. (1937): “A Note on Measurement of Utility,” The Review of Economic

Studies, 4 (2), 155-161.

Sayman, S. and A. Onculer (2009): “An investigation of time inconsistency,” Management

Science, 55, 470-482.

Shapiro, J. (2010): “Discounting for You, Me, and We: Time Preference in Groups and Pairs,”

Working Paper.

Shapiro, Jesse M (2005): “Is there a daily discount rate? Evidence from the food stamp

nutrition cycle,” Journal of Public Economics, 89 (2), 303-325.

Stephen Marglin (1963): “The Social Rate of Discount and The Optimal Rate of Investment,”

The Quarterly Journal of Economics, 77.1, p95-111.

Sutter, Matthias, Martin G. Kocher, Daniela Glatzle-Ruetzler, and Stefan T. Trautmann

(2013): “Impatience and Uncertainty: Experimental Decisions Predict Adolescents’ Field

Behavior,” American Economic Review, 103 (1), 510-531.

Tanaka, Tomomi, Colin Camerer, and Quang Nguyen (2010): “Risk and time preferences:

Experimental and household data from Vietnam,” American Economic Review, 100 (1), 557-

571.

Thaler, R. (1981): “Some empirical evidence on dynamic inconsistency,” Economics Letters,

8, 201-207.

Voors, Maarten, Eleonora Nillesen, Philip Verwimp, Erwin Bulte, Robert Lensink, and Daan

van Soest (2012): “Violent Conflict and Behavior: A Field Experiment in Burundi,” American

Economic Review, 102 (2),941-964.

Warner, John and Saul Pleeter (2001): “The Personal Discount Rate: Evidence from Military

Downsizing Programs,” American Economic Review, 91 (1), 33-53.


Appendix

A1: Nonlinear Least Squares Method

Let there be N experimental subjects and P Convex Time Budget (CTB) budgets. Assume

that each subject j makes her v1tij, i = 1, 2, ..., P , decisions (individual and group both)

according to the non-linear Euler equation mentioned above but that these decisions are made

with some mean-zero, potentially correlated error. That is let

f(V,R,1d=1, k, b1, δ, β, b2) =

(b2β

1d=1δkV

b1R2 + b2β1d=1δk

).

then

v1t∗ij = f(V,R,1d=1, k, b1, δ, β, b2) + etij.

Stacking the P observations for individual j making her individual and group decisions, we

have

v1t∗j = f(V,R,1d=1, k, b1, δ, β, b2) + ej.

The vector ej is zero in expectation with variance covariance matrix Vj , a (P × P ) matrix,

allowing for arbitrary correlation in the errors eij . We stack over the N experimental subjects

to obtain

v∗1t = f(V,R,1d=1, k, b1, δ, β, b2) + e.

We assume that the terms eij may be correlated within groups (or individuals within the

same group) but that the errors are uncorrelated across groups (or individuals within the

same group), E(e′jeg) = 0 for j 6= g. And so e is zero in expectation with covariance matrix

Ω, a block diagonal (NP ×NP ) matrix of clusters, with groups covariance matrices, Vj . We

define the usual criterion function S(V,R,1d=1, k, b1, δ, β, b2) as the sum of squared residuals,

S(V,R,1d=1, k, b1, δ, β, b2) =N∑j=1

P∑i=1

(v1t∗ij − f(V,R,1d=1, k, b1, δ, β, b2))2,

and minimize S(.) using non-linear least squares with standard errors clustered on the group

level to obtain β, δ, and ( b1b2

). NLS procedures permitting the estimation of preference pa-

rameters at the aggregate or individual level are implemented in many standard econometrics

packages (in our case, Stata).

It is important to recognize the strengths and weaknesses of such an NLS preference estima-

tor. Parameters b1 and b2 can be estimated as opposed to assumed, which is an advantage. A

potential disadvantage is that the NLS estimator is not well-suited to the censored data issues


inherent to potential corner solutions. Although the NLS estimator can be adapted to account

for possible corner solutions by adapting the criterion function and making additional distri-

butional assumptions but in this paper we did not take this approach. Andreoni and Sprenger

(2012) suggest a complementary approach: taking logs of the Euler Equation allows one to

estimate the model parameters in a two-limit Tobit framework that corrects for censoring. As

expected given the total number of corner points in the data the Table 7 results based on the

method by Andreoni and Sprenger (2012) are very close to the results in Table 3.

In the Euler Equation Approach the parameters of interest can be recovered from non-linear

combinations of regression coefficients with standard errors calculated via the delta method. As

discussed in Andreoni and Sprenger (2012) one important issue to consider in estimation is the

potential presence of corner solutions. We provide estimates from two-limit tobit regressions

designed to account for the possibility that the tangency condition implied by task’s Euler

Equation does not hold with equality (Wooldridge, 2002).

As a robustness check for our main Table 3 estimates Table 7 is presented. In the table

comparing the estimates of column (1) with (4), column (2) with (5), and column (3) with

(6) one can observe that they are very close. The values of χ2(1) = under the specified null

hypothesis are also not that different signifying the estimates obtained and presented in Table

3 of the main part of the paper are robust to the problems of censoring.


Table 8: Regression Analysis

Dependent variable: Tasks Allocated to the First Day (lnv1∗it)

Euler Equation Approach NLS Approach

(1) (2) (3) (4) (5) (6)

β 0.776*** 0.813*** 0.706*** 0.782*** 0.819*** 0.712***

(0.049) (0.053) (0.057) (0.054) (0.060) (0.059)

δ 0.997*** 0.977*** 1.038*** 0.975*** 0.959*** 1.009***

(0.031) (0.031) (0.037) (0.035) (0.035) (0.040)

( b1b2

) 1.085*** 0.994*** 1.295*** 1.072*** 1.001*** 1.231***

(0.208) (0.191) (0.277) (0.233) (0.217) (0.294)

# of Obs 2196 1464 732 2196 1464 732

# of Clusters 122 122 122 122 122 122

Pse-LL -2817 -1893 -922 -1781 -1199 -578

BIC 5672 3822 1876 3585 2421 1176

H0 : β = 1

χ2(1) = 16.01 9.25 23.74 21.08 12.63 26.67

p-value= 0.000 0.002 0.000 0.000 0.000 0.000

H0 : δ = 1

χ2(1) = 0.50 1.40 0.05 0.01 0.55 1.08

p-value= 0.481 0.237 0.816 0.917 0.456 0.299

H0 : ( b1b2

) = 1

χ2(1) = 0.10 0.00 0.62 0.17 0.00 1.14

p-value= 0.757 0.997 0.431 0.682 0.973 0.286

Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at group level. χ2(1)0.05 = 3.84.

The table presents the comparison of structural estimates of intertemporal hyperbolic discounting model

using Non-linear Least Squares Regression technique and Euler Equation approach. Table also shows

the p-values of different tests for the two proposed methods. Column (1) and (4) compare and present

the combine decisions estimates of β, δ, and ( b1b2

). Column (2) and (5) captures the structural estimates

of the model for the sub sample of individual decisions only. Column (3) and (6) represents the groups’

structural estimates based on the two proposed methods.


A2: Additional Individual and Group Preference Parameters Esti-

mates

In this section of appendix we provide four tables of additional group and its individual

members’ estimates of present bias, discount rate, and cost of effort parameters. The estimates

are based on NLS.

Table 9: Group Vs. Individual Estimates

G.No (vI1v

A2

vA1 vI2)G βG (

vI1vA2

vA1 vI2)i1 βi1 (

vI1vA2

vA1 vI2)i2 βi2

1 1.08 1.10 0.55 0.97 1.13 1.173 0.02 0.28 0.02 0.49 0.73 3.954 0.40 0.40 0.35 0.33 0.53 0.525 1.01 1.00 1.74 1.90 0.71 0.727 1.02 1.04 1.03 1.05 1.11 1.358 1.01 1.03 0.68 0.67 0.86 0.839 5.68 0.81 0.97 0.52 3.88 0.3710 0.26 0.29 1.00 1.20 0.11 0.1111 0.35 0.36 0.09 0.09 0.11 0.5212 0.66 0.75 0.26 0.34 0.35 0.3013 1.54 1.57 0.41 0.43 0.98 0.9914 0.29 0.29 0.60 0.59 0.32 0.3415 0.43 0.95 0.41 0.92 0.21 0.2116 0.33 0.37 0.10 0.10 0.07 0.0717 0.89 0.86 1.06 1.07 1.00 1.0018 0.98 0.97 0.66 0.84 0.97 0.9719 0.90 0.90 1.00 0.90 2.46 2.4920 0.91 0.91 0.64 0.66 0.68 0.7023 0.87 0.90 0.70 0.96 1.07 1.0624 1.01 1.00 0.82 0.80 0.98 0.9725 0.78 0.77 1.99 2.08 0.83 0.8226 1.06 1.10 1.37 1.35 1.06 1.1027 0.42 5.32 0.23 2.79 0.23 2.7928 0.85 0.85 0.93 0.95 1.11 1.1529 1.00 1.00 0.20 0.19 1.00 1.0030 1.00 1.17 1.26 0.42 0.14 0.1531 0.90 0.86 0.48 0.47 1.29 1.3332 1.07 1.10 0.98 1.08 1.00 1.05

Notes: Tabulates the mean ratio of average subsequent/immediate (I) ( v1v2

) to advance (A) ( v1v2

) for effortand corresponding time preference estimates for groups and its individual members.



G.No (vI1v

A2

vA1 vI2)G βG (

vI1vA2

vA1 vI2)i1 βi1 (

vI1vA2

vA1 vI2)i2 βi2

33 0.84 0.83 0.78 0.79 0.78 0.7934 0.86 0.91 0.86 0.78 0.87 0.7135 0.99 1.05 1.29 1.88 1.00 1.0736 0.79 0.78 0.87 0.87 1.25 1.2837 0.87 0.82 1.15 1.15 1.14 1.1439 1.00 1.27 0.88 0.93 0.88 0.9340 1.18 1.21 1.00 1.00 1.31 1.3241 1.00 1.00 1.00 1.00 0.87 0.9843 1.06 1.08 1.01 1.01 6.11 6.3044 0.83 0.86 1.89 2.22 0.57 0.5545 0.54 0.54 1.01 1.01 1.01 1.0146 0.94 0.96 1.01 1.00 0.94 0.9647 0.83 0.80 0.66 0.64 0.85 0.8549 0.94 0.92 1.15 1.13 0.35 0.3451 0.75 0.77 1.00 1.00 1.02 0.8352 1.01 1.03 0.81 0.83 0.08 0.0853 4.07 4.18 2.48 2.80 1.56 1.8254 0.91 0.92 0.98 0.99 0.69 0.7055 0.53 0.53 0.16 0.26 0.66 0.6657 1.13 1.14 0.92 0.91 1.13 1.1458 1.55 1.60 1.26 1.29 0.82 0.8059 0.58 0.67 0.85 0.85 1.01 1.0160 0.73 0.65 3.67 3.80 0.62 0.6061 0.98 0.99 1.00 1.00 0.76 0.8362 1.01 1.00 1.01 1.02 1.00 0.9963 0.44 0.43 1.09 1.12 1.06 1.0864 0.91 0.93 1.17 1.19 1.63 1.6665 0.75 0.76 0.62 0.61 1.00 1.0066 0.94 0.90 0.75 0.86 2.53 2.7867 0.99 0.97 4.72 4.96 0.16 0.1668 0.34 0.34 1.20 1.25 0.35 0.3369 1.10 1.10 0.72 0.70 1.12 1.1470 0.57 0.57 0.18 0.18 1.00 0.1871 0.04 0.14 0.86 0.80 1.35 1.1172 1.41 1.43 1.20 1.21 1.00 1.2179 1.04 1.03 0.99 0.99 0.99 0.9980 0.98 0.97 0.96 0.95 0.83 0.82






G.No (vI1v

A2

vA1 vI2)G βG (

vI1vA2

vA1 vI2)i1 βi1 (

vI1vA2

vA1 vI2)i2 βi2

81 0.59 0.58 1.00 0.86 1.00 1.0082 0.77 0.51 0.10 0.10 1.09 1.1084 0.03 0.03 0.03 0.03 0.03 0.0388 1.01 1.01 1.03 1.02 1.11 1.1489 0.69 0.67 4.54 3.93 1.69 1.7190 4.02 4.49 1.02 1.02 1.92 1.9592 0.58 0.57 1.24 1.23 3.33 2.5693 0.69 0.70 0.90 1.15 0.70 0.7195 0.40 0.40 0.50 0.52 1.00 0.5296 1.22 1.23 2.55 2.82 0.98 0.9897 2.85 2.98 0.87 0.84 1.83 1.87102 0.31 0.29 0.64 0.63 1.24 1.29103 0.89 1.00 0.15 1.94 0.82 0.81104 1.03 1.13 2.40 2.35 0.90 0.89105 0.77 0.75 0.65 0.65 0.97 0.99106 0.93 0.94 0.60 0.59 0.98 0.98107 0.84 0.82 1.63 1.65 1.19 1.19108 0.27 0.26 0.75 1.20 1.10 1.11109 0.27 0.29 0.83 0.85 0.89 0.88111 1.83 2.04 1.29 1.32 1.07 1.25112 0.78 1.01 0.88 0.92 0.82 0.89113 0.98 0.79 0.05 0.05 1.42 1.24114 0.09 0.08 2.06 2.22 0.19 0.12115 0.50 0.71 0.12 0.12 0.55 0.62116 0.99 1.02 0.53 0.65 0.92 0.95117 0.39 0.32 0.50 0.42 0.31 0.18118 0.20 0.18 0.38 0.34 0.49 0.48119 0.96 0.68 1.46 1.58 1.02 1.02122 1.46 1.48 1.40 1.36 0.58 0.61124 0.08 0.08 1.76 1.72 0.06 0.15125 0.07 0.07 0.12 0.13 0.23 0.24126 1.06 1.05 0.95 0.94 0.99 0.98128 0.73 0.72 0.70 0.67 0.74 0.72129 0.75 0.76 0.87 0.87 0.15 0.10130 0.27 0.25 0.27 0.25 0.34 0.38131 1.11 1.11 0.83 0.81 0.99 1.04133 0.64 0.19 1.91 1.90 2.18 2.26






G.No (vI1v

A2

vA1 vI2)G βG (

vI1vA2

vA1 vI2)i1 βi1 (

vI1vA2

vA1 vI2)i2 βi2

135 0.99 0.99 1.53 1.54 1.00 1.00137 1.13 1.12 0.81 0.83 0.99 1.04138 0.44 0.43 0.77 0.76 0.71 0.70139 0.43 0.43 2.34 2.44 1.03 0.99143 1.18 0.92 0.98 0.98 1.06 1.10144 0.66 0.67 0.89 0.87 1.69 1.70146 0.22 0.19 1.30 1.36 0.29 0.28147 0.99 0.97 0.97 0.95 0.99 0.99148 0.13 0.89 1.50 0.94 1.13 1.07149 1.18 1.22 0.80 0.87 1.80 1.36150 0.58 0.58 0.33 0.32 1.10 1.10151 0.95 0.93 0.93 0.89 1.01 0.98152 1.06 1.03 1.06 1.04 4.07 4.30153 10.51 10.87 15.75 15.09 16.17 11.50155 1.36 2.32 0.96 1.04 0.40 0.40156 3.48 3.14 1.00 1.00 16.60 13.04157 0.82 0.63 1.05 1.00 0.84 0.67158 0.02 0.02 0.18 0.17 0.41 0.39161 0.30 0.49 1.00 0.49 0.90 0.97162 1.22 1.53 1.09 0.95 2.13 0.90





A3: Additional Individual vs. Group Analysis

Table 13: Additional Individual vs. Group Regression Analysis


(1) (2) (3) (4) (5) (6) (7) (8)

|∆(Age (in years))| -0.06 -0.06

(0.14) (0.14)

|∆(Gender)| -0.10 -0.20

(0.41) (0.45)

|∆(Outside Class Study Hrs)| -0.12 -0.12

(0.08) (0.07)

|∆(On Campus Job)| -0.61 -0.82

(0.37) (0.51)

|∆(Family Income in Log)| -0.17 -0.18

(0.21) (0.24)

|∆(Past Savings Acc.)| 0.22 0.19

(0.49) (0.92)

|∆(Curr.Savings Acc.)| -0.59 1.05

(0.37) (0.76)

Constant 1.08*** 1.04*** 1.212*** 1.06*** 1.11*** 0.97*** 1.14*** 1.42***

(0.19) (0.18) (0.23) (0.12) (0.18) (0.10) (0.17) (0.38)

# of Groups 122 120 120 120 110 122 122 110

R2 0.00 0.00 0.01 0.01 0.00 0.00 0.01 0.06

Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at group level. The table presents the estimates of other

important demographic variables’ differences on groups’ present biased estimated variable.

Table 12 shows the association of additional individual characteristics (mentioned in the demo-

graphic section) with the group estimated present bias parameter. The results signify the fact

that beyond discount factor heterogeneity there is no association between group procrastina-

tion tendency and differences in groups’ individual members’ characteristics per say. Table 13

of this section presents the robustness test of the point estimates obtained in Table 7. Control-

ling for the other important demographic variables mentioned in the empirical literature, the

point estimates of variation in individual member discount factors and Acquaintance Duration

between them, remain the same. F stats also indicate that given there are no variations in in-

dividual members’ discount rates and in effort cost parameters the groups allocation decisions

would represent time consistent pattern even if the individuals are exponential discounters

with no Acquaintance Duration.


Table 14: Individual (IND) vs. Group Regression Analysis


(1) (2) (3) (4) (5) (6)

|∆(δIND

)| -3.84** -3.47*** -3.76*** -3.86*** -3.26*** -3.53***

(1.28) (1.18) (1.12) (1.20) (1.16) (1.27)

|∆(( b1b2

)IND

)| 0.20 0.24 0.23 0.32 0.23 0.37

(0.22) (0.30) (0.22) (0.25) (0.22) (0.25)

(βmin ≈ 1)IND

(= 1) 0.16 0.11 0.16 0.11

(0.30) (0.45) (0.36) (0.54)

(βmax ≈ 1)IND

(= 1) -0.23 -0.34 -0.06 -0.04

(0.13) (0.23) (0.12) (0.16)

Acquaintance Duration 0.63** 0.63*** 0.64** 0.66**

(0.27) (0.23) (0.27) (0.27)

|∆(δIND

)|×Acq Duration -6.29** -6.64** -6.37* -6.60**

(2.70) (2.89) (2.75) (2.94)

Constant 1.23*** 1.54*** 1.13*** 1.29*** 1.05*** 1.11

(0.21) (0.49) (0.14) (0.30) (0.14) (0.30)

Control for Demographic Variables No Yes No Yes No Yes

# of Groups 122 110 122 110 122 110

R2 0.04 0.10 0.23 0.28 0.34 0.39

H0 : Constant = 1

p-value= 0.26 0.35 0.71

p-value= 0.27 0.32 0.69

Notes: *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at group level. Column (1) again represents

the formal test of Jackson and Yariv (2015)’s main hypothesis. Column (5) represents the robustness of the results

obtained in column(1) controlling for differences in demographic variables mentioned in Table 13. Column (3), and

(5) is the same as in Table 7, and column (4), and (6) represents the corresponding robustness check of the results

obtained.


A4: Experiment Protocol

Instructions

Thank you for participating in our experiment. We will begin shortly.

Eligibility:

To be in this study, you need to meet the following criteria:

• You must be willing to participate for three consecutive Fridays. Participation will require

your presence on specific days as outlined.

• You will need at least one hour and at max three hours on Friday 13th March, Friday

20th March and Friday 27th March.

Informed Consent:

Placed in front of you is an informed consent form to protect your rights as a subject. Please

read it. If you choose not to participate in the study you are free to leave at this point,

deciding to leave later would seriously harm our resources allocated to this study. If you have

any questions, we can address those now. We will collect the forms after the main points of

the study are discussed.

Anonymity:

Your anonymity in this study is assured. All the information that we acquired, will only be

used for the purpose of communication with you. After the study, your email information will

be destroyed and will not be connected to your responses in the experiment.

Venue:

• Venue for Friday 20th March will be the same.

• For Friday 27th March, you dont have to be present physically. You can work from

anywhere remotely given that you have internet connection.

Rules:

• Please turn your own cell phones off.

• If you have a question at any point, just raise your hand.

• There will be a short survey once we are finished with the instructions.

• During the process of reviewing your answers in your survey if we find your responses in

violation of any of the instructions, you might get removed from the experiment.


• You will get Rs.500 as participation fee. Participation means showing up on first two

Fridays.

• If you complete the assigned tasks on all required days of participation as instructed, a

completion payment of Rs.1500 will be provided.

• You may receive additional earnings during the experiment if you participate in potential

survey games.

• If you choose to end your participation before the completion of the experiment, please

report this to study administrators at the mentioned email address.

• All payments will be made on 1st April in IGC office room 161. You will return the

phones given to you for experiment purposes to IGC to receive this payment.

Task:

In this study there is only one task. This task will be completed over time. Some portion

of the tasks may be completed sooner, and some portion of the task can be completed later

depending on your choices and chance.

This task will consist of taking specific number of pictures of books through cell phones.

Remember your phone has a unique IMEI number. Once you take a picture, you need to

upload the picture using the application on the phone. Make sure your pictures are clear

and the numbers are legible. If the numbers are not legible, they will not be counted. Some

portion of the tasks may be completed on second Friday, and some portion of the task can

be completed on third Friday. You will practice the working of phone application before the

actual task starts.

Task Rates:

The allocation decisions across two weeks depend on the task rate. The task rate will vary

across your decisions. On the target-setting page of the application (installed in the cell phones

given to you), every slider bar corresponds to a specific task rate. For example in the first

slider bar the task rate is 1:0.8, such that every task you allocate to second Friday reduces

the number of tasks allocated to third Friday by 0.8. For simplicity, the task rates will always

be represented as 1:X, and you will be fully informed of the value of X when making your

decisions.

The Experiment Timeline:

Before explaining the activities to be done on each Friday you need to have an overall picture

of the timeline.

First Friday (13th March 2015):


• Physical Presence • Alloca&on Decision Day 1

Friday 13th March

• Physical Presence • Alloca&on Decision Day 2 • Task Comple:on Day 1

Friday 20th March

• Task Comple:on Day 2

Friday 27th March

Figure 5: Timeline

Notes: Three weeks experimental timeline figure provided to all participants

• First of all, all of the Subjects will be required to fill out different survey forms.

• After the Survey forms have been completed and collected, you will be asked to make a

series of 3 decisions for task distribution as an individual.

• Once you have made decisions for individual task distribution, again you will be required

to make 3 decisions and distribute the task as a group.

• Keep in mind that your decision today is for the task you will be doing on 20th and 27th.

This applies to both Individual and Group decisions.

• In each decision you are free to allocate your tasks as you choose.

Second Friday (20th March 2015):

• During our second session here in the very same venue, again you will be asked to make

3 Decisions (both individually and as a group) as you did during first session.

• By this time we will have 12 decisions from every subject. (3 Individual + 3 Group on

13th and 3 Individual + 3 Group on 20th)

• Exactly one of your 12 total decisions will be implemented randomly.

• We will discuss how this allocation decision is chosen during our training session.

• We refer to this allocation decision as the ”decision-that-counts.” The tasks you allocated

yourself for 20th and 27th in the decision-that-counts must be completed.

• If you do not complete the tasks according to the decision-that-counts, you will not

receive completion amount of Rs. 1500 and will only receive participation fee of Rs. 500.

• In order for your tasks on second or third Friday to be counted, they must be completed

between 9:30 pm and 10:30 pm of that Friday.

• Surveys will be conducted which will give you a chance to earn more money.


Third Friday (27th March 2015):

• You will have to complete your tasks for this day according to your decision-that-counts

• You can do this remotely from anywhere.

How we will choose the Decision-That-Counts:

The process of selecting the Decision-that-Counts is simple probability. Three stages to deter-

mine the decision-that-counts:

1. First you will be allocated either 13th March Decisions or 20th March Decisions according

to a 20% and 80% chance respectively.

2. Once you have been allocated to a specific date (13th or 20th) either you will be given

an Individual Task or a Group Task with 33% and 66% chance respectively.

3. After both the steps given above are complete, you will receive one of the 3 decisions you

made for that specific date and specific task type with equal chance.

EACH DECISION COULD BE THE DECISION-THAT-COUNTS SO TREAT EACH DE-

CISION AS IF IT WAS THE ONE DETERMINING YOUR TASKS.

Short Survey: Please answer the following questions:

1. How many weeks do we require you to participate?

2. In which of the three weeks are you asked to participate remotely and not come to this

venue?

3. What is the percent chance that one of your 20th March allocations will be implemented?

4. If you face a 1:2 task rate for allocations between Weeks 2 and 3, every task you allocate

to Week 2 decreases by how many number of tasks you allocate to Week 3?

Documents

Fickle Groups: An Empirical Assessment of Time ......I am grateful to Michael Callen (Harvard Kennedy School), Aprajit Mahajan (UC Berkeley), and Charlie Sprenger (UC San Diego) for