Temptation: Immediacy and certainty*

J. Lucas Reddinger†
arXiv:2407.14955v3 [econ.GN] 7 Aug 2024

7 August 2024

Abstract Is an option especially tempting when it is both immediate and certain? I test the
effect of risk on the present-bias factor given quasi-hyperbolic discounting. In my experiment,
workers allocate about thirty to fifty minutes of real-effort tasks between two weeks. I study
dynamic consistency by comparing choices made two days in advance of the workday with
choices made when work is imminent. My novel design permits estimation of present bias
using a decision with a consequence that is both immediate and certain. I find greater present
bias when the consequence is certain. This finding has implications for any economic decision
involving a present-biased decision-maker, including labor contracting and consumer good
pricing. Finally I offer a methodological remedy for experimental economists.

JEL Codes C91, D80, D90

Keywords present bias, dynamic inconsistency, quasi-hyperbolic discounting, time prefer-

ences, risk preferences, immediacy effect, certainty effect, experimental economics

∗I give special thanks to Gary Charness, Ignacio Esponda, and Erik Eyster. I also thank David Freeman, Peter
Kuhn, and participants at University of California, Santa Barbara, seminars and Economic Science Association
meetings. Reddinger (2019) pre-registered this study with the American Economic Association as AEARCTR-
0004651. The University of California, Santa Barbara, Human Subjects Committee exempted my protocol 56-19-
0621. Reddinger (2024) provides data and source code. The Department of Economics and the Graduate Division
at the University of California, Santa Barbara, provided funding.
† Department of Economics, University of California, Santa Barbara, United States. Current affiliation: De-

partment of Economics, Purdue University, 403 West State Street, Krannert Building, Room 712, West Lafayette,
Indiana 47907-2076; [email protected].

1
1 Introduction

While risk and time preferences are fundamental to the theory of decision-making, much
remains unknown about the interplay between these two dimensions. A future prospect
is inherently risky if any circumstance may arise that precludes consumption of the conse-
quence. This implies that outcomes are obtained with certainty only if obtained without
delay. Accordingly a preference for certain outcomes results in a preference for immediate
outcomes. Conversely, the introduction of risk may especially diminish the appeal of an im-
mediate reward, relative to a delayed reward. I explore in this paper such an interaction
between immediacy and certainty with an experiment of dynamic decision-making over risky
and delayed prospects.
Present-biased preferences are likely the cause of oft-bemoaned consumer behavior, such
as the failure to meet exercise goals and the over-utilization of credit-card debt (Royer,
Stehr, and Sydnor 2015; Meier and Sprenger 2010). Firms exploit consumer present bias
and successfully extract welfare (DellaVigna and Malmendier 2004). Meanwhile, incentives,
commitment devices, and other interventions may (or may not) help consumers improve
their long-term welfare (Ashraf, Karlan, and Yin 2006; Carrera et al. 2022). A thorough
understanding of present bias assists further development of these pursuits.
My study informs labor contract design, especially those used in the modern gig economy.
For example, drivers for ride-hail companies face nearly identical decisions as the workers I
hire in my experiment. These companies carefully withhold selective information (such as
ride length or destination) when offering a gig to a driver and require commitment prior
to revealing all of these ride details (Rana 2020). Such uncertainty in a spot labor contract
potentially affects labor supply; my results suggest that a present-biased worker with a weekly
income target may procrastinate less in an environment with greater uncertainty.
My novel experimental design allows estimation of present bias for subjects making a
single decision with a certain consequence. This is in contrast to a baseline treatment that
implements an allocation choice made on a randomly-selected day and at a randomly-selected

2
intertemporal price ratio, in accord with prevailing experimental methodology.
Workers in my experiment allocate a workload between two weeks. Each worker first
makes allocation decisions two days before the first workday. Each worker then returns on
the first workday and makes identical decisions with the work being imminent. If a worker
is present-biased, she will in advance choose some allocation between the two weeks, but
then on the first workday, prefer an allocation with less work for the present day. In my
experiment, once the implementation mechanism selects a particular allocation, the worker
must complete the tasks allocated to each week to earn a substantial bonus payment.
Ultimately I find that the immediacy effect is significantly attenuated by the introduction
of risk. Specifically, the quasi-hyperbolic present-bias factor 𝛽 becomes smaller with the
elimination of risk, implying greater myopia. In my experiment, when a workload allocation
is implemented with certainty, subjects on average discount the future by a factor of 𝛽ˆ =
0.581 relative to the present. In the baseline treatment that uses prevailing experimental
methodology, I find no statistically significant present bias, with 𝛽ˆ = 1.009 when each decision
has a 10% implementation probability.
These findings underscore the importance of decision-theoretic frameworks that permit
interaction between dimensions of risk and time. The findings also suggest that studies
of tempting goods may necessitate decisions with temporally salient and certain conse-
quences; researchers should keep this in mind when designing either lab or field experi-
ments. Further, risk introduced by randomized incentive mechanisms—common experimen-
tal methodology—may require augmentation of decisions with certain consequences. I offer
such a methodological remedy.

2 Background

To model intertemporal decision-making, Samuelson (1937) introduced exponential discounted

utility (DU), which describes how an individual values utility flows (of consumption goods,
such as leisure) that occur over time. If utility flows 𝑢 (𝑐 𝑡 +𝜏 ) result from consumption 𝑐 𝑡 +𝜏 at

3
time 𝑡 + 𝜏 ∈ ℕ, given a constant discount factor 𝛿 ∈ [ 0, 1] , the model gives an intertemporal
value at time 𝑡 of

∑︁
𝑈𝑡DU = 𝛿 𝜏 𝑢 (𝑐𝑡 +𝜏 ). (1)
𝜏=0

A decision-maker with this value function will make dynamically consistent choices, assuming
that the felicity function 𝑢 is time-invariant (Halevy 2015).
To capture a preference for immediate utility, Laibson (1997) introduces the present-bias
factor 𝛽 ≥ 0 to discount all future utility flows contra present utility. The resultant quasi-
hyperbolic discounted utility (QHD) has an intertemporal value at time 𝑡 of

∑︁
𝑈𝑡QHD = 𝑢 (𝑐𝑡 ) + 𝛽 𝛿 𝜏 𝑢 (𝑐𝑡 +𝜏 ). (2)
𝜏=1

𝛽 < 1 describes a preference for immediacy, also referred to as present bias. This is an
example of diminishing sensitivity to delay—an individual is more impatient regarding a delay
in felicity that happens immediately relative to a delay that occurs in the future. Meanwhile,
some individuals may exhibit future bias, with 𝛽 > 1.
A decision-maker with 𝛽 ≠ 1 will make dynamically inconsistent choices. A decision-maker
with 𝛽 < 1 (𝛽 > 1) will continually revise consumption plans to achieve greater (lesser)
felicity in the present moment relative to her prior plans.

2.1 Diminishing sensitivity to risk and delay

Let a simple gamble (𝑥 ◦ 𝑝) be a prospect that obtains 𝑥 with probability 𝑝 . Given the
independence axiom of expected utility theory (EUT), a preference relation is maintained if
prospect probabilities are multiplied by a common ratio (Machina 1982). Consider

Menu A : 𝑎 = ( 1 ◦ 0.9) or 𝑎 ′ = ( 2 ◦ 0.6) ; and

Menu B : 𝑏 = ( 1 ◦ 0.6) or 𝑏 ′ = ( 2 ◦ 0.4).

4
Under EUT, 𝑎 is weakly preferred to 𝑎 ′ if and only if 𝑏 is weakly preferred to 𝑏 ′ .
Allais (1953) presented empirical violations of this result (see also Kahneman and Tversky
1979). The common ratio effect describes a preference reversal in which a decision-maker is
indifferent between two prospects, but when the prospect probabilities are scaled down by a
common ratio, she then strictly prefers the riskier option (e.g., 𝑎 ∼ 𝑎 ′ and 𝑏 ≺ 𝑏 ′ ).
The certainty effect is a special case of the common ratio effect when one prospect obtains
with probability one. For example, consider

Menu C : 𝑐 = ( 3 ◦ 1.0) or 𝑐 ′ = ( 4 ◦ 0.8) ; and

Menu R : 𝑟 = ( 3 ◦ 0.5) or 𝑟 ′ = ( 4 ◦ 0.4).

If a decision-maker is indifferent between 𝑐 and 𝑐 ′ but strictly prefers 𝑟 ′ over 𝑟 , she may
simply possess diminishing sensitivity to risk as described by the common ratio effect, or she
may have a disproportionate preference for a certain outcome.
Prelec and Loewenstein (1991) note equivalent results regarding time delay using dis-
counted utility. Following Halevy (2008), let us simply interpret 𝛿 in equation (1) as a failure
risk imposed by a unit-time delay that precludes consumption of the consequence (e.g., 𝛿
might be one’s probability of death in every time period). Under DU, a decision-maker
weakly prefers one intertemporal consumption plan to another if and only if this preference
is maintained with an additional arbitrary time delay.
Consider a daily survival probability of 0.8. Let us reinterpret the previous menus as

Menu C̃ : 𝐶 = 3 now or 𝐶 ′ = 4 in 1 day; and

Menu R̃ : 𝑅 = 3 in 3 days or 𝑅 ′ = 4 in 4 days.1

The common difference effect describes a preference reversal in which a decision-maker is

indifferent between two intertemporal consumption plans, but when an arbitrary time delay

1With approximation, 0.5 is the probability of surviving ln 0.5/ln 0.8 ≈ 3 days, and 0.4 is the probability of
surviving ln 0.4/ln 0.8 ≈ 4 days.

5
is added to each, she then becomes more patient (e.g., 𝐶 ∼ 𝐶 ′ and 𝑅 ≺ 𝑅 ′ ).
The immediacy effect is a special case of the common difference effect, when only immediate
consumption varies between plans. In equation (2), QHD describes an immediacy effect if
𝛽 < 1. Chakraborty, Halevy, and Saito (2020) fully characterize the relationship between
the immediacy effect (including under QHD) and the certainty effect.2
In this sense, a decision-maker only obtains a prospect with certainty when also obtained
without delay. Any time delay plausibly eliminates a certainty effect, and similarly any risk
plausibly eliminates an immediacy effect. I endeavor to experimentally test the significance
of this interaction.

2.2 Evidence of risk moderating present bias

At least three studies have shown that risk moderates present bias using hypothetical or
nearly-hypothetical monetary incentives.
Keren and Roelofsma (1995) provided the first evidence of risk moderating present bias.
In this between-subject full-factorial experiment, subjects chose between $50 and $55 with a
four-week delay. When prizes obtain with certainty, 82% of subjects prefer $50 immediately
over $55 in four weeks, while only 37% of subjects prefer $50 in twenty-six weeks over $55
in thirty weeks, thereby demonstrating present bias at certainty. When prizes obtain with
probability one-half, 39% of subjects prefer $50 immediately over $55 in four weeks, and 33%
of subjects prefer $50 in twenty-six weeks over $55 in thirty weeks, failing to show significant
present bias.
Weber and Chapman (2005) confirm these findings, again with hypothetical monetary
stakes. Baucells and Heukamp (2010) also find that risk moderates present bias with highly-
diluted monetary incentives, implementing only three of 3,757 decisions.3
While these prior studies find that risk moderates present bias, they may drastically under-
estimate the intensity of the effect due to the methodologies used. Hypothetical decisions
2Epper and Fehr-Duda (2018), Baucells and Heukamp (2010), and Green and Myerson (2004) also explore
this relationship.
3Three of 221 subjects are selected, each of whom has one of their seventeen decisions implemented.

6
lack incentive, relying solely on framing and contingent reasoning. Similarly, extremely
low probability of implementation may dilute the stated probability of the prospects due to
isolation failure (which I discuss in section 2.4).
Models of present bias are often used to study self-control failure of visceral urges (Cheung,
Tymula, and Wang 2021), which are plausibly best elicited with an immediate and certain
consequence. For example, many studies of sequential games find costly punishment more
prevalent upon eliciting a direct-response action instead of a conditional strategy (Brandts
and Charness 2011), perhaps due to a preference for exacting revenge.
My study is the first to use truly immediate and certain consequences in studying present
bias. I avoid concerns associated with hypothetical decisions, long-shot implementation, and
contingent reasoning, thereby establishing an ideal method for capturing present bias.

2.3 Empirical estimates of present bias

While many studies have used monetary rewards to measure present bias, Andreoni and
Sprenger (2012) pioneered the “convex time budget” (CTB) methodology, which elicits
monetary-prize allocations between two time periods at various interest rates, thereby al-
lowing risk preferences and QHD parameters to be estimated jointly.⁴ However, monetary
earnings may not adequately capture consumption utility in the absence of liquidity con-
straints and decision isolation. Augenblick, Niederle, and Sprenger (2015) address this
concern with CTB decisions in which individuals allocate real-effort tasks across time (other
studies have used alternative primary rewards, such as food).
The meta-analysis by Imai, Rutter, and Camerer (2021) finds no evidence of present bias
in monetary rewards, while finding a mean bias-corrected present-bias factor 𝛽 between
0.90 and 0.99 in real-effort tasks. While the specific value depends on the particular bias
correction, present bias is also highly context-dependent.⁵

⁴See Cheung, Tymula, and Wang (2021) for a nice review of estimates of present bias using various method-
ologies. For more on these other methodologies used to elicit time and risk preferences, see Andersen et al. (2008),
Cheung (2016, 2020), and Abdellaoui et al. (2013).
⁵While Imai, Rutter, and Camerer (2021) restrict their focus to twenty-eight studies that use the CTB methodol-

7
2.4 Decision framing, isolation, and implementation

A typical subject in an economics experiment makes many decisions. Historically many ex-
periments implement many or all decisions, but this method can yield data rife with wealth
effects, hedging, and other confounds (Charness, Gneezy, and Halladay 2016). Consequently
most experiments now implement one randomly-selected decision, thus avoiding such com-
plementarity between outcomes (Azrieli, Chambers, and Healy 2018). Yet implementing one
decision at random is not a panacea; many subjects still fail to isolate each decision.⁶
Non-expected utility rationalizes isolation failure when a subject views a set of decisions
as comprising a single optimization problem. For example, given a Holt and Laury (2002)
choice list, a subject is often able to secure a certain outcome by choosing every safe option.
Freeman, Halevy, and Kneeland (2019) find evidence of the certainty effect when comparing
pairwise choices to choice-list data. Freeman and Mayraz (2019) find evidence of the Allais
paradox with the effect independent of the mechanism used.
Freeman and Mayraz (2019) find that presentation has the largest impact on isolation.
Brown and Healy (2018) display decisions separately and reclaim incentive compatibility.

3 Experimental design

The present study compares present bias between a risky consequence and a certain conse-
quence. I implement one decision to avoid complementarity (see section 2.4), which must be
implemented with certainty in the respective treatment. To obtain useful choice data from
a single implemented decision, I use the CTB methodology (see section 2.3). To induce an
immediate (primary) reward, I ask subjects to allocate a budget of real-effort tasks between
two weeks.

ogy, Cheung, Tymula, and Wang (2021) offer a meta-analysis which includes studies that use other methodologies
such as the joint-elicitation methodology (Andersen et al. 2008). Xueting Wang has informed me that a forth-
coming revision of Cheung, Tymula, and Wang (2021) will present results that are qualitatively consistent with
those of Imai, Rutter, and Camerer (2021).
⁶See Starmer and Sugden (1991), Beattie and Loomes (1997), Cubitt, Starmer, and Sugden (1998), and Cox,
Sadiraj, and Schmidt (2015).

8
 The experiment consists of three sessions: Monday (day zero), Wednesday (day two), and
the following Wednesday (day nine). Every subject earns $1.50 per session, which must
be completed between noon and midnight. I immediately disqualify subjects who miss a
deadline. Every subject that completes all three sessions earns a $5 bonus.
Each session begins with ten mandatory tasks, providing salient experience and a fixed
baseline effort-level on each day. Each task asks the subject to count the number of zero digits
in a sixteen-digit binary string and enter this count into an adjacent text field (figure 1a).
The subject must remedy any incorrect response for successful submission.
Following the mandatory tasks on day zero and day two, each subject chooses an allocation
of 360 tasks between day two and day nine at each of five substitution rates (the 360-task
budget has day-two value). Each decision is presented individually to elicit a tentative choice
(figure 1b), which are then juxtaposed on a single page (figure 1c) for any adjustment.
Accordingly, let 𝑒𝑖,𝑑
𝑡 denote effort chosen at rate 𝑅 on decision-day 𝑑 to be expended on
𝑖
day two
workday 𝑡 . For example, 𝑒 4,0 is the effort chosen at rate 𝑅4 on day zero to be worked on
day two. Each subject faces the constraint

day two day nine

𝑒𝑖,𝑑 + 𝑅𝑖 𝑒𝑖,𝑑 = 360, for each 𝑅𝑖 ∈ R B ⟨1.25, 0.75, 1, 1.5, 0.5⟩ and 𝑑 ∈ 𝐷 B {0, 2} .

Some rates 𝑅𝑖 (which can also be interpreted as productivity ratios or gross interest rates)
entail substantial income effects. For example, at 𝑅4 = 1.5, a subject may choose to delay
all 360 tasks on day two to only 240 tasks on day nine. Yet if a subject delays all 360 tasks
at 𝑅5 = 0.5, she would need to complete 720 tasks on day nine. Randomly-selected subjects
receive a rate sequence in reverse order.

3.1 Treatments

I implement a 2 × 2 factorial between-subject design. I inform all subjects that a decision

from either day zero or day two will be selected with equal probability (figure 1d).
The baseline treatment implements one of the ten decisions with uniform probability.

9
PRACTICE MODE The correct answers are already filled in to save you time. PRACTICE MODE You will not have to work these tasks.

Complete 10 required rows of counting Split workload between Wed, Oct 30 and Wed, Nov 6
Please count the number of zeros (“0”) on each line and enter it in the box. You're making five decisions on how to split the workload for Wed, Oct 30. You'll make five more similar decisions
on that day.
Each row will be marked correct or incorrect. You must correct errors before submission.
A coin flip will determine whether a decision made today or a decision made on Wed, Oct 30 will be selected to
Row No. String Count (“0”) actually matter.

1 8 One of today's five decisions may be randomly selected to actually split your workload.

2 12 The odds of each decision being the decision-that-matters are 10%.

3 8 Trade-off Wed, Oct 30 Wed, Nov 6

4 7 1 to 0.5 360 rows 0 rows

5 10 1 to 0.75 274 rows 115 rows

6 8 1 to 1 139 rows 221 rows

7 10 1 to 1.25 40 rows 256 rows

8 8 1 to 1.5 0 rows 240 rows

9 5 Please review your choices and make any final changes.

10 8 Finalize

Check responses and save

(a) Task interface (c) Allocation interface, presented juxtaposed

PRACTICE MODE You will not have to work these tasks. PRACTICE MODE

Split workload between Wed, Oct 30 and Wed, Nov 6 How today's decisions are used
Choose how you want to split your workload of 360 rows of counting (in addition to the required 10 rows per You made decisions about splitting work between this Wednesday and next Wednesday.
workday).
You will make similar decisions again Wednesday. One day will be selected for its decisions to actually matter.
In this scenario, working 1 more row next week reduces work by 1 row(s) this week.
Sun, Oct 27 Mon, Oct 28 (today) Tue, Oct 29 Wed, Oct 30 Thu, Oct 31 Fri, Nov 1 Sat, Nov 2
You're making five decisions on how to split the workload for Wed, Oct 30. You'll make five more similar decisions
on that day.
Decisions made Decisions made

A coin flip will determine whether a decision made today or a decision made on Wed, Oct 30 will be selected to Sun, Nov 3 Mon, Nov 4 Tue, Nov 5 Wed, Nov 6 Thu, Nov 7 Fri, Nov 8 Sat, Nov 9
actually matter.

One of today's five decisions may be randomly selected to actually split your workload.

The odds of this decision being the decision-that-matters are 10%. You just made five decisions about how to split work between these
days
Wed, Oct 30 Drag slider handle to adjust choice. Wed, Nov 6
Choice No. Trade-off Wed, Oct 30 Wed, Nov 6

139 rows 221 rows 1 1 to 0.5 360 rows 0 rows

Today's choices
2 1 to 0.75 235 rows 167 rows
3 1 to 1 139 rows 221 rows
4 1 to 1.25 52 rows 247 rows
Try moving the slider around to see how this trade-off rate splits your workload. 5 1 to 1.5 0 rows 240 rows

If this choice were selected to actually matter, your work schedule would be:
You will make five similar decisions Wednesday
Sun, Oct 27 Mon, Oct 28 (today) Tue, Oct 29 Wed, Oct 30 Thu, Oct 31 Fri, Nov 1 Sat, Nov 2
10 rows required Choice No. Trade-off Wed, Oct 30 Wed, Nov 6

+ 139 rows chosen 1 1 to 0.5 x rows x rows

Wednesday's choices
2 1 to 0.75 x rows x rows
Sun, Nov 3 Mon, Nov 4 Tue, Nov 5 Wed, Nov 6 Thu, Nov 7 Fri, Nov 8 Sat, Nov 9 3 1 to 1 x rows x rows
4 1 to 1.25 x rows x rows
10 rows required 5 1 to 1.5 x rows x rows
+ 221 rows chosen

You will be able to adjust this decision before finalizing it. After you make decisions Wednesday, a coin-toss will select which
day's decisions are used
Continue
Choice No. Trade-off Wed, Oct 30 Wed, Nov 6 Choice No. Trade-off Wed, Oct 30 Wed, Nov 6
1 1 to 0.5 360 rows 0 rows 1 1 to 0.5 x rows x rows

Today's choices Wednesday's choices

2 1 to 0.75 235 rows 167 rows 2 1 to 0.75 x rows x rows
3 1 to 1 139 rows 221 rows 3 1 to 1 x rows x rows
4 1 to 1.25 52 rows 247 rows 4 1 to 1.25 x rows x rows
5 1 to 1.5 0 rows 240 rows 5 1 to 1.5 x rows x rows

Reveal

(b) Allocation interface, presented separately (d) Day selection mechanism interface

Figure 1: Experimental interface

10
 Table 1: Probability of decision implementation
Decision chosen on
Treatment day 𝑑 = 0 day 𝑑 = 2
Baseline 1/10 1/10
Certain Day 1/10 1/5
Certain Rate 1/2 1/2
Certain Rate, Certain Day 1/2 1
Note: Probabilities of implementation of the effort allocation
choice 𝑒 2,𝑑 (chosen on decision-day 𝑑 at rate 𝑅2 = 1.25).

Subjects with this treatment have uncertainty regarding the day and the rate selected. Each
decision in this Risky Rate, Risky Day treatment thus has a 10% implementation probability.
The Certain Rate (CR) treatment dimension eliminates risk regarding the rate to be im-
plemented. In this treatment, subjects are informed that 𝑅2 = 1.25 will certainly be im-
plemented; decisions for all prices 𝑅𝑖 ≠ 1.25 are hypothetical, which I exclude from my
analysis.
The Certain Day (CD) treatment dimension eliminates risk regarding the day from which a
decision is selected. I inform subjects in this treatment that I will reveal the randomly-selected
day before their day-two decisions. Accordingly, the day to be implemented is risky for all
subjects on day zero, but certain for subjects with CD treatment on day two. Half of the
subjects with this treatment learn that their day-two decisions are hypothetical; I exclude all
hypothetical decisions from analysis and compensate by doubling the sample size of the CD
dimension.⁷
Table 1 shows decision implementation probability by treatment cell 𝑇 .
My primary interest is the interaction of Certain Rate and Certain Day treatments. On day
zero subjects know that their choice at 𝑅2 = 1.25 made on either day zero or made on day two
will be selected with certainty. Then on day two, prior to making a decision, subjects learn

⁷I considered this alternate design: If day zero is selected, inform after day-two decisions; if day two is
selected, inform before day-two decisions. I rejected this design because subjects would lack complete prior
information about the timing of the resolution of risk.

11
from which day a decision will be selected. Subjects who learn day two is selected thus make
a decision on day two that is certainly implemented. That is, these subjects choose their
impending same-day effort level, knowing this choice will be implemented with certainty,
with the tasks due imminently. I hypothesize that present bias is more pronounced under
certainty than under risk.

3.2 Interface

I carefully designed the interface to bolster subjects’ understanding of the choice process
and the implementation mechanism. The interface guides every subject through a complete
practice round at the beginning of each session before the real decisions and tasks.

Day selection The interface shows each subject a list of their practice choices made on
each decision-day (figure 1d). Upon clicking the button to proceed, the page visualizes
random selection between the two decision-days by alternately highlighting the lists in quick
succession before the highlight settles on one day as being selected. Every subject simulates
two practice coin-flips: the first trial selects the alternate decision-day, then the second selects
the present day. The remainder of the practice round uses the choices made in the present
session.

Rate selection The interface next shows each subject the five practice choices made on
the present day, arranged in a table similar to the juxtaposed allocation page (figure 1c).
Subjects with Certain Rate treatment see row four permanently highlighted and a reminder
that this choice will be certainly implemented. Other subjects see no highlight at first, but
upon clicking the button to proceed, a roulette-wheel sequence highlights each row quickly
in succession. After traversing the table twice, the highlight settles on a randomly-selected
decision.
With a practice allocation selected, subjects view a practice task interface that requests the
corresponding amount of work to be completed on the present day. Subjects then exit the

12
practice round and begin an identical sequence with real decisions and tasks.

4 Model and methodology

Assuming quasi-hyperbolic discounting, within-day power-function effort costs, and back-

ground effort of 𝜔 , on each decision-day 𝑑 = 0 and 𝑑 = 2 the decision-maker optimizes

 𝛼  𝛼
𝟙(𝑑=0 )
min 𝛽
𝑡
𝑒𝑖,𝑑 + 𝜔 + 𝛽𝛿 𝑒𝑖,𝑑 + 𝜔 , subject to 𝑒𝑖2 + 𝑅𝑖 𝑒𝑖9 = 360,
2 7 9
(3)
𝑒𝑖,𝑑

𝑡 for each price ratio 𝑅 ∈ { 0.5, 0.75, 1, 1.25, 1.5 } and workday 𝑡 ∈ { 2, 9 } .
choosing effort 𝑒𝑖,𝑑 𝑖

This model uses 𝛿 as a daily discount factor, while 𝛽 discounts future-day effort. Assuming
the independence axiom, the resultant intertemporal Euler equation is:

2
! 𝛼 −1
𝑒𝑖,𝑑 +𝜔 𝛽 𝟙(𝑑=2 ) 𝛿 7
9
= (4)
𝑒𝑖,𝑑 +𝜔 𝑅𝑖

(Lawrance 1991). Logarithms linearize this equation as

2
𝑒𝑖,𝑑 +𝜔 ln 𝛿 −1 ln 𝛽
ln = 7+ ln 𝑅𝑖 + 𝟙(𝑑 = 2) . (5)
9
𝑒𝑖,𝑑 + 𝜔 𝛼 −1 𝛼 −1 𝛼 −1
| {z } |{z} |{z} |{z} | {z }
C𝐸𝑖,𝑑 C𝜃 delay C𝜃 lnrate C𝜃 present C𝟙( pr )

Let us define the variables as shown under braces in equation (5) to simplify notation. An
additive error term produces an estimatable reduced-form, with 𝑠 indexing subjects:

𝐸𝑖,𝑑,𝑠 = 𝜃 delay 7 + 𝜃 lnrate ln 𝑅𝑖 + 𝜃 present 𝟙( pr)𝑑 + 𝜀𝑖,𝑑,𝑠 . (6)

Let 𝟙( cr) indicate Certain Rate treatment and 𝟙( cd) Certain Day treatment. We interact the
full-factorial of these with the present-workday indicator, 𝟙(𝑑 = 2) , to obtain an estimatable

13
pooled reduced-form regression model:

𝐸𝑖,𝑑,𝑠 = 𝜃 delay 7 + 𝜃 lnrate ln 𝑅𝑖 + 𝜃 present 𝟙(𝑑 = 2)𝑑 + 𝜃 cr 𝟙( cr)𝑠 𝟙(𝑑 = 2)𝑑

+ 𝜃 cd 𝟙( cd)𝑠 𝟙(𝑑 = 2)𝑑 + 𝜃 cr,cd 𝟙( cr)𝑠 𝟙( cd)𝑠 𝟙(𝑑 = 2)𝑑 + 𝜀𝑖,𝑑,𝑠 . (7)

This specification allows recovery of 𝛽𝑇 that varies by treatment cell 𝑇 (the supplement offers
details).

4.1 Hypotheses

My primary hypothesis is that an interaction exists between the immediacy effect and the
certainty effect. That is, present bias at certainty differs from present bias with risk. The
baseline treatment with risk regarding the rate and risk regarding the decision-day is standard
in the literature, here with each decision having an implementation probability of 1/10. When
treated with Certain Rate and Certain Day, the day-two choice for 𝑅2 = 1.25 is implemented
with certainty (probability of one).

Hypothesis 1 Present-bias is more intense under implementation certainty (with both Certain
Rate and Certain Day treatment) than when the decision involves both types of risk (decision-day
and rate both unrealized), in which each decision has an implementation probability of 1/10:
𝛽 cr,cd < 𝛽 .

For completeness I further hypothesize that present bias at certainty differs from present
bias with any uncertainty—that is, with only one dimension of risk.

Hypothesis 2 Present-bias is more intense under implementation certainty than when the de-
cision involves rate risk but has decision-day certainty (implementation probability of 1/5):
𝛽 cr,cd < 𝛽 cd .

Hypothesis 3 Present-bias is more intense under implementation certainty than when the de-
cision involves decision-day risk but has rate certainty (implementation probability of 1/2):
𝛽 cr,cd < 𝛽 cr .

14
 I do not hypothesize further how the type of risk may matter. For example, controlling for
the implementation probability, perhaps risk regarding the rate is most influential, perhaps
driven by the income effect of the price ratios. Research regarding types of risk and underlying
mechanisms is left to future research.

4.2 Statistical methodology

Each subject allocates 360 tasks (in day-two valuation) between day two and day nine at
various price ratios. Each subject must also complete ten mandatory real-effort tasks on each
day. A subject might most prefer a negative effort allocation to a workday (that is, a net gain
of leisure on that day), which my environment does not facilitate: the environment bounds
2
elicited day-two effort such that 𝑒𝑖,𝑑 ∈ [0, 360] for all rates 𝑅𝑖 on each decision-day 𝑑 . A
two-limit Tobit model accommodates this censoring.
To estimate the model with a power cost function 𝑐 (𝑒) B (𝑒 + 𝜔) 𝛼 , we must specify some
background effort 𝜔 > 0. The primary analysis will use 𝜔 = 10 as the background effort,
corresponding to the mandatory daily tasks. Subjects may perform other tasks throughout
the day that we might wish to include in 𝜔 ; my supplement offers results that demonstrate
robustness to various background effort levels.

4.3 Identification of present bias

As represented by the factor 𝛽 , present bias is identified from a two-day window. On Monday,
day zero, I assume that subjects view both Wednesday, day two, and the next Wednesday, day
nine, as part of the future. Then on day two, I assume that subjects view that same day as
the present and continue to view day nine as the future.
One could reasonably argue that Monday and Wednesday of the same week may both feel
relatively present, while the following week may feel relatively distant. This would imply that
present bias would be better identified from a week-long delay, as in Augenblick, Niederle,
and Sprenger (2015). However, this is an empirical question, and Augenblick (2018) studies

15
exactly how present bias varies with short delays. Using similar real-effort tasks, he finds
that present bias quickly diminishes within three days, with two days capturing most present
bias. In the present study, the use of a two-day window will yield conservative estimates of 𝛽
(biased upward). A week-long delay might more accurately identify present bias, but likely
at the cost of greater attrition. Regardless, in the present study I simply intend to use average
treatment effects to test my hypotheses.

4.4 Identification of discounting

The daily discount factor 𝛿 exponentially discounts the future. In this environment, we
identify 𝛿 (jointly with 𝛼 ) using variation in the rate 𝑅𝑖 .
Suppose that marginal cost of effort is constant within a day, so that the effort-cost convexity
parameter 𝛼 = 1. Then the ratio 𝛿 7 /𝑅𝑖 determines how a decision-maker allocates her
workload between day two and day nine. If 𝛿 7 /𝑅𝑖 = 1, she is indifferent to how the workload
is split; otherwise she will allocate the entire workload to a single day. For example, if she
discounts the future (𝛿 < 1) but she can trade day-two and day-nine work one-for-one (when
𝑅𝑖 = 1), she will choose to do all of the work on day nine.
Instead, assume that the decision-maker has an increasing marginal cost of within-day
effort, so that 𝛼 > 1. Then if 𝑅𝑖 = 1 and 𝛿 = 1, she would divide the workload evenly between
day two and day nine. This is because she values smoothing effort between workdays, since
additional within-day effort becomes more costly. Then as either the rate 𝑅𝑖 or the discount
factor 𝛿 changes, the decision-maker will choose a different workload split between day two
and day nine, balancing the benefit of smoothing effort against the inferior rate.
Because 𝛿 7 /𝑅𝑖 and effort-cost convexity 𝛼 jointly explain allocation between workdays, 𝛼
and 𝛿 are jointly identified by rate 𝑅𝑖 variation.

16
4.5 Between- and within-subject identification

My novel experimental design and identification strategy permits estimation of the present-
bias factor 𝛽 with some subjects making only two incentivized decisions, as in the interaction
of Certain Rate and Certain Day treatments. This is possible because effort-cost convexity 𝛼
and the discount factor 𝛿 are estimated with the pooled regression, relying on variation in rate
𝑅𝑖 from subjects with Risky Rate treatment. Meanwhile, 𝛽𝑇 is estimated between-subjects for
each treatment cell 𝑇 .⁸
Other researchers who wish to evaluate present bias under certainty need not implement
my full-factorial design. My design will identify 𝛼 , 𝛿 , and 𝛽𝑇 with only two treatment cells: a
cell with Risky Rate and a cell with Certain Rate and Certain Day. The cell with Risky Rate can
use either Risky Day or Certain Day; however the latter requires a larger sample to achieve
the same statistical power if the study excludes hypothetical decisions and avoids deception.
For my present study I implement a full-factorial design to investigate each dimension.

5 Results

I recruited subjects from an online piece-rate labor marketplace with an equivalent median
hourly wage under $5 and jobs that commonly involve transcribing invoices or tagging pho-
tographs (Newman 2019).⁹ I described my “multi-day counting project” as three sessions
with a combined 30–50 minutes of tasks which paid $1.50 for each session and a $5 bonus for
completing all three. Given that the experiment involves tasks similar to those typical of the
marketplace, it is a framed field experiment (Harrison and List 2004).
My instructions provided sample tasks, explained the task allocation process, and stressed
the three dates of participation: Monday 28 October 2019 (day zero), Wednesday 30 October
(day two), and Wednesday 6 November (day nine). Consenting subjects answered an eight-

⁸While within-subject estimation of the present-bias factor is possible, note that each point-estimate 𝛽𝑠 would
rely on only two observations, 𝑒 1,0 and 𝑒 1,2 , per subject 𝑠 given Certain Rate treatment.
⁹On Amazon Mechanical Turk I sampled residents of the United States and Canada with at least 1,000 jobs
completed with 98% approval.

17
question comprehension survey which paid $1.50 regardless of the responses (the supplement
provides all experimental instruments).
Of the 389 comprehension survey submissions, 220 provided informed consent and only
correct responses; I enrolled these subjects in my experiment. Of these 220 subjects, 206
(93.6%) enrolled in and completed day-zero decisions. From this first session to the last,
sample attrition was only 26 of 206 subjects (12.6%).1⁰ The median subject completed a
combined 340 tasks in 36 minutes (with quartiles 𝑞 1 = 28 and 𝑞 3 = 47 minutes).

5.1 Descriptive results

While log-effort-ratio is the correctly specified choice variable given the model in equation (3),
2
let us consider a more intuitive outcome: day-two effort-share 𝜑 B 𝑒𝑖,𝑑 /360. Because the
Certain Price treatment only incentivizes 𝑅2 = 1.25, I only analyze choice data at this rate.
Note that because day-nine effort is more productive than day-two effort at this rate, effort is
split evenly between workdays when 𝑒 22,𝑑 = 𝑒 29,𝑑 = 160 and thus 𝜑 = 160/360 = 0.44.
Figure 2 offers histograms of effort-share choices by treatment. Given an effort-share
choice in advance of the workday (day zero), a smaller choice on the workday itself (day two)
suggests present bias. Thus this graph illustrates present bias if the outlined bars shift to the
left of the solid bars.
We see a striking feature at 0.0–0.10: in treatments with at least one certainty treatment,
many subjects choose a day-two allocation much different than their day-zero allocation. On
day two, with work imminent, they choose to exert 0.0–0.10 of effort on day two. Compare
the filled bar against the outlined bar for the 0.0–0.10 bin in each treatment; more subjects
choose an allocation in this bin on day two (when the work is imminent) relative to day zero
(when the work is distant) with either certainty treatment.
This simple descriptive graph intuitively suggests evidence of present bias in at least some
treatments. We now turn to regression analysis to make use of all incentivized choice data.
1⁰On day two, 192 subjects (93.2%) returned and completed that day’s decisions. One of each subject’s ten
decisions was implemented, upon which 188 subjects (97.9%) completed the selected day-two effort. Finally, on
day nine, 180 subjects (95.7%) returned and completed the session, thus earning the completion bonus.

18
Figure 2: Histograms of effort-share chosen for day two at 𝑅2 = 1.25 for each treatment

Treatment: Risky rate, risky day Treatment: Risky rate, certain day
.6

.6
Fraction of choices

Fraction of choices
.4

.4
.2

.2
0

0
0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1
Effort share for day two Effort share for day two

Treatment: Certain rate, risky day Treatment: Certain rate, certain day
.6

.6
Fraction of choices

Fraction of choices
.4

.4
.2

.2
0

0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1
Effort share for day two Effort share for day two

Choices made on
Day zero Day two

19
5.2 Regression results

We now consider the regression results as presented in table 2 and figure 3. We notably find
no present bias in the baseline treatment. Meanwhile, relative to the baseline treatment with
risk, present bias under certainty is vastly different economically. With a point-estimate of
0.58 under certainty, subjects value the present 1.7 times as much as they value the future.
We reject the null hypothesis that 𝛽 cr,cd = 𝛽 at 𝑝 < 0.001. This provides clear evidence that
the introduction of a substantial amount of risk significantly moderates present bias.
We also find that addition of risk in the rate dimension alone also drastically moderates
present bias relative to the baseline, rejecting 𝛽 cr,cd = 𝛽 cd with 𝑝 = 0.005. However we do
not find a similar result in comparing certainty to the addition of risk in the decision-day
dimension, failing to reject that 𝛽 cr,cd = 𝛽 cr with 𝑝 = 0.458. Recall that with Certain Rate
treatment, we rely on only two observations per subject; otherwise we have ten observations
per subject. Indeed the standard errors for 𝛽 with Certain Rate treatment are roughly double
those in the other treatments, suggesting that a larger sample may reveal significance.
Beyond concerns of statistical power, we can conjecture that the implementation probability
plays an important role (figure 3). When making a day-two decision with Certain Rate and
Risky Day treatments, subjects know that their present decision will be selected by one side
of a coin flip. This might make the decision sufficiently salient to preserve a high degree of
present bias, especially with only a single incentivized decision on each day. Regardless, the
testing of such hypotheses and underlying mechanisms is left to future work.

5.3 Additional considerations

Attrition bias
While sample attrition (12.6%) was remarkably low for an online experiment across ten days,
we should look for evidence of selective attrition. For example, Certain Rate treatment might
have lower attrition as it guarantees an income effect, whereas the Risky Rate treatment does
not in expectation.

20
 Table 2: Regression results
𝑝 -value of 𝜒12 test that parameter equals
Param. Estim. Std. Err. 1 𝛽 𝛽 cd 𝛽 cr
𝛽 cr,cd 0.581 ( 0.108) < 0.001 < 0.001 0.005 0.458
𝛽 cr 0.679 ( 0.109) 0.003 0.011 0.046
𝛽 cd 0.921 ( 0.057) 0.166 0.327
𝛽 1.009 ( 0.055) 0.873
𝛿 0.986 ( 0.004) 0.001
𝛼 1.282 ( 0.045) < 0.001
Note: 897 observations (161 left- and 95 right-censored) from 180 subjects. Robust standard
errors in parentheses are clustered on subject using a two-limit Tobit model.

Figure 3: Regression estimates of 𝛽𝑇

Risky rate,
1.1 risky day

1.0 1.009 𝛽

0.921 𝛽 cd
0.9
Risky rate, Certain rate,
certain day risky day
0.8
Certain rate,
certain day
𝛽ˆ𝑇

0.7
0.679 𝛽 cr

0.6 𝑝 = 0.327 0.581 𝛽 cr,cd

𝑝 = 0.046
0.5
𝑝 = 0.011
𝑝 = 0.458
0.4 𝑝 = 0.005
𝑝 < 0.001
0.3
0.0 0.2 0.4 0.6 0.8 1.0
Implementation probability 𝑝

21
 Only four subjects completed day-two decisions but did not complete the implemented
day-two effort level. Two of these made only hypothetical decisions on day two and are
thus excluded from the analysis. Both remaining subjects had Certain Rate treatment. We
conclude that attrition during day two was orthogonal to rate resolution.
Attrition between day two and day nine of eight subjects was highly balanced across
treatments and rate selection.

Effort-cost curvature
We reject the hypothesis that 𝛼 ≥ 1 with 𝑝 < 0.001, satisfying the second-order condition for
equation (3). The results are robust to background effort 𝜔 of greater orders of magnitude,
appropriate for having already worked prior to the sessions (see the supplement).

6 Conclusion

This study of dynamic inconsistency in real-effort provision finds that risk diminishes the
intensity of present bias. This includes uncertainty that arises from random-implementation
mechanisms popular among experimental economists. My experiment varies the implemen-
tation mechanism, thereby altering the probability of decision implementation. The novel
design permits pooled estimation of the present-bias factor in each of four treatment cells,
including one that implements a single decision with certainty.
The effect of certainty on present bias is striking. Under certainty I estimate 𝛽ˆcr,cd = 0.581,
while the baseline treatment finds no significant present bias with a point-estimate of 𝛽ˆ =
1.009. These results present a remarkable treatment effect: risk significantly moderates
present bias.
While most other studies find present bias in real-effort, some do not (Imai, Rutter, and
Camerer 2021). The replication study of Augenblick, Niederle, and Sprenger (2015) is similar
to my baseline treatment, using the same implementation mechanism, one-week delay, and
similar interest rates. They estimate 𝛽ˆ = 0.892 with 𝑝 = 0.05 for University of California
students transcribing blurry Greek letters. My subjects, being workers in an online market-

22
place, may have substituted effort in my session with effort in other jobs. Nevertheless the
treatment effect suggests that the implementation mechanism affects present bias.
Experiments that seek an accurate point-estimate of the present-bias factor should include a
decision with substantial immediate and certain consequences. If complementarities between
consequences do not pose a serious concern, the experiment might reasonably implement
multiple such decisions.
My findings underscore the importance of unifying theories of time and risk, notably
Chakraborty, Halevy, and Saito (2020) (see section 2.1 and footnote 2). Conversely, in
testing decision-theoretic models, researchers should mind their incentive mechanisms and
use decisions implemented with certainty when appropriate.
Empirical work on tempting goods may require decisions with salient and certain conse-
quences, a potentially-critical design element for any study employing experimental methods.
Such work might study models of self-control, the effectiveness of commitment devices, or
any application that depends on present-biased preferences.
Uncertainty may interact with non-stationary time preferences, leading to different behav-
ior in strategic interactions. For example, in labor contracts firms may exploit present bias
with (un)certainty regarding compensation, effective productivity, or job duration.
Clearly the field of behavioral economics has much yet to learn about present bias, temp-
tation, and related interventions. Continual improvement of experimental methodology will
aid this pursuit.

References
Abdellaoui, Mohammed, Han Bleichrodt, Olivier l’Haridon, and Corina Paraschiv. 2013. “Is
there one unifying concept of utility? An experimental comparison of utility under risk
and utility over time.” Management Science 59 (9): 2153–2169.
Allais, Maurice. 1953. “Le comportement de l’homme rationnel devant le risque: Critique
des postulats et axiomes de l’école américaine.” Econometrica 21 (4): 503–546.

23
Andersen, Steffen, Glenn W. Harrison, Morten I. Lau, and E. Elisabet Rutström. 2008. “Elic-
iting risk and time preferences.” Econometrica 76 (3): 583–618.
Andreoni, James, and Charles Sprenger. 2012. “Estimating time preferences from convex
budgets.” American Economic Review 102 (7): 3333–56.
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 (2): 635–672.
Augenblick, Ned. 2018. “Short-term time discounting of unpleasant tasks.” http://faculty.
haas.berkeley.edu/ned/Augenblick_ShortTermDiscounting.pdf.
Augenblick, Ned, Muriel Niederle, and Charles Sprenger. 2015. “Working over time: Dy-
namic inconsistency in real effort tasks.” Quarterly Journal of Economics 130 (3): 1067–
1115.
Azrieli, Yaron, Christopher P. Chambers, and Paul J. Healy. 2018. “Incentives in experi-
ments: A theoretical analysis.” Journal of Political Economy 126 (4): 1472–1503.
Baucells, Manel, and Franz H. Heukamp. 2010. “Common ratio using delay.” Theory and
Decision 68 (1–2): 149–158.
Beattie, Jane, and Graham Loomes. 1997. “The impact of incentives upon risky choice ex-
periments.” Journal of Risk and Uncertainty 14 (2): 155–168.
Brandts, Jordi, and Gary Charness. 2011. “The strategy versus the direct-response method:
a first survey of experimental comparisons.” Experimental Economics 14 (3): 375–398.
Brown, Alexander L., and Paul J. Healy. 2018. “Separated decisions.” European Economic
Review 101 (January): 20–34.
Carrera, Mariana, Heather Royer, Mark Stehr, Justin Sydnor, and Dmitry Taubinsky. 2022.
“Who chooses commitment? Evidence and welfare implications.” The Review of Eco-
nomic Studies 89 (3): 1205–1244.
Chakraborty, Anujit, Yoram Halevy, and Kota Saito. 2020. “The relation between behavior
under risk and over time.” American Economic Review: Insights 2 (1): 1–16.
Charness, Gary, Uri Gneezy, and Brianna Halladay. 2016. “Experimental methods: Pay one
or pay all.” Journal of Economic Behavior & Organization 131A (November): 141–150.
Cheung, Stephen L. 2016. “Recent developments in the experimental elicitation of time
preference.” Journal of Behavioral and Experimental Finance 11:1–8.
. 2020. “Eliciting utility curvature in time preference.” Experimental Economics 23
(2): 493–525.

24
Cheung, Stephen L., Agnieszka Tymula, and Xueting Wang. 2021. Quasi-hyperbolic present
bias: A meta-analysis. Technical report 14625. IZA Institute of Labor Economics.
Cox, James C., Vjollca Sadiraj, and Ulrich Schmidt. 2015. “Paradoxes and mechanisms for
choice under risk.” Experimental Economics 18 (2): 215–250.
Cubitt, Robin P., Chris Starmer, and Robert Sugden. 1998. “On the validity of the random
lottery incentive system.” Experimental Economics 1 (2): 115–131.
DellaVigna, Stefano, and Ulrike Malmendier. 2004. “Contract design and self-control: The-
ory and evidence.” The Quarterly Journal of Economics 119 (2): 353–402.
Epper, Thomas, and Helga Fehr-Duda. 2018. “The missing link: Unifying risk taking and
time discounting.” Working Paper 96. University of Zurich, Department of Economics.
https://www.zora.uzh.ch/id/eprint/66623/.
Freeman, David J., Yoram Halevy, and Terri Kneeland. 2019. “Eliciting risk preferences us-
ing choice lists.” Quantitative Economics 10 (1): 217–237.
Freeman, David J., and Guy Mayraz. 2019. “Why choice lists increase risk taking.” Experi-
mental Economics 22 (1): 131–154.
Green, Leonard, and Joel Myerson. 2004. “A discounting framework for choice with delayed
and probabilistic rewards.” Psychological Bulletin 130 (5): 769–792.
Halevy, Yoram. 2008. “Strotz meets Allais: Diminishing impatience and the certainty effect.”
American Economic Review 98 (3): 1145–62.
. 2015. “Time consistency: Stationarity and time invariance.” Econometrica 83 (1):
335–352.
Harrison, Glenn W., and John A. List. 2004. “Field experiments.” Journal of Economic Litera-
ture 42 (4): 1009–1055.
Holt, Charles A., and Susan K. Laury. 2002. “Risk aversion and incentive effects.” American
Economic Review 92 (5): 1644–1655.
Imai, Taisuke, Tom A. Rutter, and Colin F. Camerer. 2021. “Meta-analysis of present-bias
estimation using convex time budgets.” Economic Journal 131 (636): 1788–1814.
Kahneman, Daniel, and Amos Tversky. 1979. “Prospect Theory: An analysis of decision un-
der risk.” Econometrica 47 (2): 263–292.
Keren, Gideon, and Peter Roelofsma. 1995. “Immediacy and certainty in intertemporal choice.”
Organizational Behavior and Human Decision Processes 63 (3): 287–297.
Laibson, David. 1997. “Golden eggs and hyperbolic discounting.” Quarterly Journal of Eco-
nomics 112 (2): 443–478.

25
Lawrance, Emily C. 1991. “Poverty and the rate of time preference: evidence from panel
data.” Journal of Political Economy 99 (1): 54–77.
Machina, Mark J. 1982. “‘Expected utility’ analysis without the independence axiom.” Econo-
metrica 50 (2): 277–323.
Meier, Stephan, and Charles Sprenger. 2010. “Present-biased preferences and credit card
borrowing.” American Economic Journal: Applied Economics 2 (1): 193–210.
Newman, Andy. 2019. “I found work on an Amazon website. I made 97 cents an hour.” The
New York Times (November 15, 2019). https://www.nytimes.com/interactive/2019/11/
15/nyregion/amazon-mechanical-turk.html.
Prelec, Drazen, and George Loewenstein. 1991. “Decision making over time and under un-
certainty: A common approach.” Management Science 37 (7): 770–786.
Rana, Preetika. 2020. “Uber retools California fares in response to gig economy law; Move
is latest change by ride-hailing giant to bolster case that drivers are independent work-
ers.” Wall Street Journal (Online) (January). https : / / www . wsj . com / articles / uber -
retools-california-fares-in-response-to-gig-economy-law-11578527202.
Reddinger, J. Lucas. 2019. “Temptation: Immediacy and certainty.” AEA RCT Registry. https
://doi.org/10.1257/rct.4651.
. 2024. (Temptation). https://doi.org/10.17605/OSF.IO/ZE3WJ.
Royer, Heather, Mark Stehr, and Justin Sydnor. 2015. “Incentives, commitments, and habit
formation in exercise: evidence from a field experiment with workers at a fortune-500
company.” American Economic Journal: Applied Economics 7 (3): 51–84.
Samuelson, Paul A. 1937. “A note on measurement of utility.” Review of Economic Studies 4
(2): 155–161.
Starmer, Chris, and Robert Sugden. 1991. “Does the random-lottery incentive system elicit
true preferences? An experimental investigation.” American Economic Review 81 (4):
971–978.
Weber, Bethany, and Gretchen Chapman. 2005. “The combined effects of risk and time on
choice: Does uncertainty eliminate the immediacy effect? Does delay eliminate the cer-
tainty effect?” Organizational Behavior and Human Decision Processes 96 (2): 104–118.

26
 Temptation: Immediacy and certainty
Supplement

J. Lucas Reddinger
7 August 2024
arXiv:2407.14955v3 [econ.GN] 7 Aug 2024

Contents
1 Additional results 1

2 Experimental instruments 1

3 Parameter recovery 2

1 Additional results
• Table 1: Subjects by treatment
• Table 2: Regression results with various background effort
• Table 3: Estimated covariance matrix for the primary regression

2 Experimental instruments
• Table 4: Experimental timeline
• Table 5: Comprehension test

The interface included animations to convey the random selection procedure to subjects. Only one
slide from each animation is included in this document.

• Figures 1 to 6: Qualification session instructions

• Figures 7 to 9: Day zero session instructions
• Figure 10: Required tasks
• Figure 11: Task allocation, separate
• Figure 12: Task allocation, juxtaposed
• Figure 13: Decision-day selection
• Figure 14: Rate selection
• Figure 15: Example of implemented tasks

1
 Table 1: Subject counts by treatment
Treatment Subject count on day
1 2
No. Day Rate Selects day Rate order Zero Two Nine
1 Risky Risky Two Original 8 8 8
2 Risky Certain Two Original 10 10 10
3 Risky Risky Zero Original 9 8 7
4 Risky Certain Zero Original 8 7 7
5 Certain Risky Two Original 17 16 15
6 Certain Certain Two Original 17 16 15
7 Certain Risky Zero Original 18 16 16
8 Certain Certain Zero Original 15 15 13
9 Risky Risky Two Reversed 9 8 7
10 Risky Certain Two Reversed 8 8 7
11 Risky Risky Zero Reversed 9 8 8
12 Risky Certain Zero Reversed 9 9 8
13 Certain Risky Two Reversed 17 15 14
14 Certain Certain Two Reversed 17 16 15
15 Certain Risky Zero Reversed 18 16 15
16 Certain Certain Zero Reversed 17 16 15
Total subject count by day 206 192 180
1 2
The day from which a decision was selected for implementation. Decisions are presented in
the order of R or its reverse.

3 Parameter recovery
Parameters in the model of the main text are recovered as follow:
𝜃 present 𝜃 present + 𝜃 cr
𝛽 = exp , 𝛽 cr = exp ,
−𝜃 lnrate −𝜃 lnrate
𝜃 present + 𝜃 cd 𝜃 present + 𝜃 cr + 𝜃 cd + 𝜃 cr,cd
𝛽 cd = exp , 𝛽 cr,cd = exp ,
−𝜃 lnrate −𝜃 lnrate
𝜃 delay −1
𝛿 = exp , and 𝛼 = 1 − 𝜃 lnrate .
−𝜃 lnrate

2
 Table 2: Regression results with various background effort
Background effort, 𝜔
10 1200 2400 4800
Equivalent duration of 𝜔 † 1 minute 2 hours 4 hours 8 hours
𝛽 1.009 0.984 0.982 0.981
( 0.055) ( 0.068) ( 0.069) ( 0.069)
𝛽 cd 0.921 0.884 0.882 0.880
( 0.057) ( 0.072) ( 0.074) ( 0.075)
𝛽 cr 0.679 0.665 0.662 0.661
( 0.109) ( 0.113) ( 0.114) ( 0.114)
𝛽 cr,cd 0.581 0.569 0.566 0.565
( 0.108) ( 0.113) ( 0.114) ( 0.114)
𝛿 0.986 0.987 0.987 0.987
( 0.004) ( 0.005) ( 0.005) ( 0.005)
𝛼 1.282 5.048 8.668 15.898
( 0.045) ( 0.688) ( 1.317) ( 2.575)
𝑝 -value of 𝜒 12 test:
𝛽 = 𝛽 cd 0.327 0.366 0.369 0.371
𝛽 = 𝛽 cr 0.011 0.019 0.019 0.020
𝛽 = 𝛽 cr,cd 0.001 0.002 0.002 0.002
𝛽 cd = 𝛽 cr 0.046 0.084 0.086 0.087
𝛽 cr,cd = 𝛽 cd 0.005 0.012 0.012 0.012
𝛽 cr,cd = 𝛽 cr 0.458 0.481 0.481 0.481
𝛽=1 0.873 0.808 0.795 0.788
𝛽 cd = 1 0.166 0.109 0.108 0.108
𝛽 cr = 1 0.003 0.003 0.003 0.003
𝛽 cr,cd = 1 < 0.001 < 0.001 < 0.001 < 0.001
𝛿=1 0.001 0.014 0.017 0.019
𝛼 =1 < 0.001 < 0.001 < 0.001 < 0.001
Note: 897 observations (161 left- and 95 right-censored) from 180 subjects. Robust standard errors in
parentheses are clustered on subject using a two-limit Tobit model. Excludes subjects who did not com-
†
plete all sessions. The ex ante specification uses background effort 𝜔 = 10. The median duration of 10
real-effort tasks is 1 minute.

Table 3: Estimated covariance matrix for the primary regression with 𝜔 = 10

𝛽 𝛽 cd 𝛽 cr 𝛽 cr,cd 𝛿 𝛼
𝛽 0.00303 −0.00086 −0.00099 −0.00090 0.00006 0.00014
𝛽 cd −0.00086 0.00325 0.00020 0.00027 −0.00002 −0.00044
𝛽 cr −0.00099 0.00020 0.01183 0.00297 0.00011 −0.00244
𝛽 cr,cd −0.00090 0.00027 0.00297 0.01173 0.00012 −0.00263
𝛿 0.00006 −0.00002 0.00011 0.00012 0.00002 −0.00008
𝛼 0.00014 −0.00044 −0.00244 −0.00263 −0.00008 0.00206

3
 Table 4: Experimental timeline
Day zero “Qualification HIT” Payment of $1.50 within twenty-four hours
1. Instructions
2. Consent
3. Comprehension test
A subject qualifies for the next session if and only if all comprehension answers are correct.

Day zero “Monday’s HIT” Payment of $1.50 within twenty-four hours

1. Instructions
2. Practice: Ten mandatory tasks that would need to be completed
3. Practice: Effort allocation between day two and day nine, presented separately
4. Practice: Effort allocation between day two and day nine, presented juxtaposed
5. Practice: How today’s decisions are used (resolution of decision-day risk)
6. Practice: How today’s decisions are used (resolution of rate risk)
7. Practice: View implemented tasks that would need to be completed
8. Complete the ten mandatory tasks
9. Effort allocation between day two and day nine, presented separately
10. Effort allocation between day two and day nine, presented juxtaposed
A subject qualifies for the next session if and only if all parts of this session are completed.

Day two “This Wednesday’s HIT” Payment of $1.50 within twenty-four hours
1. Instructions
2. Practice: Ten mandatory tasks that would need to be completed
3. Practice: Effort allocation between day two and day nine, presented separately
4. Practice: Effort allocation between day two and day nine, presented juxtaposed
5. Practice: How today’s decisions are used (resolution of decision-day risk)
6. Practice: How today’s decisions are used (resolution of rate risk)
7. Practice: View implemented tasks that would need to be completed
8. Complete the ten mandatory tasks
9. Certain Day treatment only: One day is selected for implementation
10. Effort allocation between day two and day nine, presented separately
11. Effort allocation between day two and day nine, presented juxtaposed
12. Risky Day treatment only: One day is selected for implementation
13. Risky Rate treatment only: One rate is selected for implementation
14. Complete the implemented tasks for today
A subject qualifies for the next session if and only if all parts of this session are completed.

Day nine “Next Wednesday’s HIT” Payment of $6.50 within twenty-four hours
1. Instructions
2. Complete the ten mandatory tasks
3. Complete the implemented tasks for today
Note: In the labor marketplace used, HIT is common nomenclature for a single job.

4
 Table 5: Comprehension test
1. On which of the following days do you plan on completing a HIT for this project?
One checkbox per date:
Sun, Oct 27; Mon, Oct 28; Tue, Oct 29; Wed, Oct 30; Thu, Oct 31; Fri, Nov 1;
Sat, Nov 2; Sun, Nov 3; Mon, Nov 4; Tue, Nov 5; Wed, Nov 6; Thu, Nov 7;
Fri, Nov 8; Sat, Nov 9
2. After you receive your qualification today, will there be another HIT to complete today?
Radio buttons: Yes; No
3. What happens if you make counting errors in the task?
Radio buttons:
The HIT might be rejected.
I must start the entire HIT over from the beginning.
I will be told which rows have errors, then I’ll correct the errors.
4. Suppose you complete the first HIT later today and you also complete the HIT on this Wednes-
day. However, you do not complete the HIT next Wednesday. How much will you earn in total
from this study? (Do not include earnings from this qualification HIT.)
Text input formatted as USD currency
5. How many HITs in total must you complete to earn the bonus? (Do not count this qualification
HIT.)
Text input
6. On how many days total (including today) must you complete a HIT to earn the bonus? (Do
not count this qualification HIT.)
Text input
7. How much will you earn in total by fully participating in this study, including the bonus? (Do
not include earnings from this qualification HIT.)
Text input formatted as USD currency
8. Each day’s HIT will definitely be available during what times? Specify a time range using
Pacific Time:
Selection menu:
Beginning at: 07:00 am; 08:00 am; 09:00 am; 10:00 am; 11:00 am; 12:00 pm;
01:00 pm; 02:00 pm; 03:00 pm; 04:00 pm; 05:00 pm; 06:00 pm
Ending at: 12:00 pm; 1:00 pm; 2:00 pm; 3:00 pm; 4:00 pm; 5:00 pm; 6:00 pm;
7:00 pm; 8:00 pm; 9:00 pm; 10:00 pm; 11:00 pm; 12:00 am (midnight);
01:00 am; 02:00 am
Note: In the labor marketplace used, HIT is common nomenclature for a single job.

5
Introduction
Hello! I'm [researcher name], a [researcher position] at [researcher institution].

I'm running this research project to complete my degree. Thank you for taking the time to read this!

I watch my email closely while my HITs are live, so feel free to email me at [researcher email].

I've tried my best to ensure that you are fairly compensated for your time. I pay all earnings within 24 hours.

If you complete the project, you are paid the same amount no matter what. The amount you have to work is partly
determined by chance, but your decisions give you control over how much you work and when you work.

Absolutely no deception is used in this experiment. All random coin tosses were generated using a computer
program that will be provided (along with the results) to the American Economic Association. My research will not be
accepted if it is not completely transparent and truthful.

You may contact the [IRB] regarding your rights and participation at [IRB email].

Qualification HIT
This HIT may qualify you for a sequence of 3 HITs for an academic research project.
This qualification HIT pays $1.50, independently of any earnings described below.
Please read these instructions carefully and answer the questionnaire at the bottom. Thank you!

What you may qualify for…

Earnings

1 more HIT today after you have been qualified; it pays $1.50
1 HIT this Wednesday; it pays $1.50
1 HIT next Wednesday; it pays $1.50
Earn a $5 bonus for completing all 3 HITs

Today's HIT requirements

This HIT should only take 5-10 minutes to complete; it pays $1.50

1. Work 10 rows of the counting task (like the example below)

2. Decide how to split a workload of 360 rows of counting between Wednesday and next Wednesday

This Wednesday's HIT requirements

This HIT may take 10-20 minutes, depending on how you split the work between the days; it pays $1.50

1. Work 10 rows of the counting task (like the example below)

2. Decide how to split a workload of 360 rows of counting between Wednesday (that same day) and next
Wednesday
3. One of your decisions from today or Wednesday is selected to actually split the workload
Figure
4. Work the counting task for the1:selected
Qualification
amount session instructions
of work for (1 of 6)
this Wednesday
6
Next Wednesday's HIT requirements

This HIT may take 10-20 minutes, depending on how you split the work between the days; it pays $1.50

1. Work 10 rows of the counting task (like the example below)

2. Work the counting task for the selected amount of work for next Wednesday

Summary

Earn $9.50 total (in addition to this $1.50 qual HIT) for about 30 to 40 minutes of work total, assuming you complete
 4. Work the counting task for the selected amount of work for this Wednesday

Next Wednesday's HIT requirements

This HIT may take 10-20 minutes, depending on how you split the work between the days; it pays $1.50

1. Work 10 rows of the counting task (like the example below)

2. Work the counting task for the selected amount of work for next Wednesday

Summary

Earn $9.50 total (in addition to this $1.50 qual HIT) for about 30 to 40 minutes of work total, assuming you complete
all 3 HITs.

Example: working 10 rows of the counting task

This is a completed example of the counting task that would be in each HIT.

Please count the number of zeros (“0”) on each line and enter it in the box.

Each row will be marked correct or incorrect. You must correct errors before submission.

Row No. String Count (“0”)

1 8

2 12

3 8

4 7

5 10

6 8

7 10

8 8

9 5

10 8

Check responses and save

Decide how to split a workload between this Wednesday and next

Wednesday
You start with a workload of 360 rows of counting. You will decide how to split up the workload between this
Wednesday and Wednesday of next week.

You decide for different trade-off scenarios. For example:

Figure 2: Qualification session instructions (2 of 6)
Working 1 more row next week reduces work by 1 row this week (a 1-to-1 trade-off)
Working 1 more row next week reduces work by 1.5 rows this week (a 1-to-1.5 trade-off)
Working 1 more row next week reduces work by 0.5 rows this week (a 1-to-0.5 trade-off)

One scenario will be selected to actually matter. 7

You will get more details and practice in the project's first HIT after you are qualified.

Complete each of 3 days' HITs for a $5 bonus, earning $9.50 total

On each day in blue below, complete one HIT for $1.50.
Once you complete all 3 HITs, you will be paid a $5 bonus.
If you fail to complete a day's HIT, you will not be able to complete further HITs, nor will you receive the bonus.
You will always be paid for HITs you have already completed. All earnings are paid within 24 hours.
Decide how to split a workload between this Wednesday and next
Wednesday
You start with a workload of 360 rows of counting. You will decide how to split up the workload between this
Wednesday and Wednesday of next week.

You decide for different trade-off scenarios. For example:

Working 1 more row next week reduces work by 1 row this week (a 1-to-1 trade-off)
Working 1 more row next week reduces work by 1.5 rows this week (a 1-to-1.5 trade-off)
Working 1 more row next week reduces work by 0.5 rows this week (a 1-to-0.5 trade-off)

One scenario will be selected to actually matter.

You will get more details and practice in the project's first HIT after you are qualified.

Complete each of 3 days' HITs for a $5 bonus, earning $9.50 total

On each day in blue below, complete one HIT for $1.50.
Once you complete all 3 HITs, you will be paid a $5 bonus.
If you fail to complete a day's HIT, you will not be able to complete further HITs, nor will you receive the bonus.
You will always be paid for HITs you have already completed. All earnings are paid within 24 hours.
Each day's HIT will be available at least between 12 p.m. and 12 a.m. (midnight), Pacific Time.
You will be qualified for today's HIT as quickly as possible, but it may take 30 minutes to an hour.
You will receive a reminder notification through mTurk on each day that you have a HIT to complete.
Please do not accept more than one HIT on the same day; it will not give you additional work or earnings.
Please only participate if you think you can complete one HIT on each of these 3 days!

Sun, Oct 27 Mon, Oct 28 (today) Tue, Oct 29 Wed, Oct 30 Thu, Oct 31 Fri, Nov 1 Sat, Nov 2
1 HIT 1 HIT
(today after qualified)
Sun, Nov 3 Mon, Nov 4 Tue, Nov 5 Wed, Nov 6 Thu, Nov 7 Fri, Nov 8 Sat, Nov 9
1 HIT

No deception is used in this experiment

This is an academic research project in the field of Economics, which widely prohibits the deception of experimental
subjects.

All information and instructions provided to you in this experiment are truthful.

If you feel anything in this experiment is deceptive, please notify the [IRB] at [IRB email].

Consent
This is an academic research project to study work decisions involving delay and uncertainty.

You may choose to quit at any time. You will still receive earnings for what you have completed. Risks are
comparable to typical computer use. There is no direct benefit to you anticipated from your participation in this
study. The data we collect will not be linked to your identity in any way.

If you have any questions about this research project, please contact [researcher] at [researcher email].
Figure 3: Qualification session instructions (3 of 6)
If you have any questions regarding your rights and participation as a research subject, please contact the [IRB
8
contact information].

Participation in research is voluntary. Clicking the button labeled “I Consent” below will indicate that you have
decided to participate as a research subject in the study described above.

I CONSENT

Qualification survey
Consent
This is an academic research project to study work decisions involving delay and uncertainty.

You may choose to quit at any time. You will still receive earnings for what you have completed. Risks are
comparable to typical computer use. There is no direct benefit to you anticipated from your participation in this
study. The data we collect will not be linked to your identity in any way.

If you have any questions about this research project, please contact [researcher] at [researcher email].

If you have any questions regarding your rights and participation as a research subject, please contact the [IRB
contact information].

Participation in research is voluntary. Clicking the button labeled “I Consent” below will indicate that you have
decided to participate as a research subject in the study described above.

I CONSENT

Qualification survey
To qualify for the study, complete this one-time survey. Earn $1.50 for a complete submission.

Please enter your worker ID:

Questions about instructions

1. On which of the following days do you plan on completing a HIT for this project?

Sun, Oct 27 Mon, Oct 28 Tue, Oct 29 Wed, Oct 30 Thu, Oct 31

Fri, Nov 1 Sat, Nov 2 Sun, Nov 3 Mon, Nov 4 Tue, Nov 5

Wed, Nov 6 Thu, Nov 7 Fri, Nov 8 Sat, Nov 9

2. After you receive your qualification today, will there be another HIT to complete today?

Yes. No.

3. What happens if you make counting errors in the task?

The HIT might be rejected. I must start the entire HIT over from the beginning.

I will be told which rows have errors, then I'll correct the errors.

4. Suppose you complete the first HIT later today and you also complete the HIT on this Wednesday. However, you
do not complete the HIT next Wednesday. How much will you earn in total from this study? (Do not include earnings
from this qualification HIT.)

5. How many HITs in total must you complete to earn the bonus? (Do not count this qualification HIT.)

6. On how many days totalFigure 4: Qualification

(including today) must yousession instructions
complete a HIT to earn(4
theofbonus?
6) (Do not count this
qualification HIT.)
9

7. How much will you earn in total by fully participating in this study, including the bonus? (Do not include earnings
from this qualification HIT.)

8. Each day's HIT will definitely be available during what times?

5. How many HITs in total must you complete to earn the bonus? (Do not count this qualification HIT.)

6. On how many days total (including today) must you complete a HIT to earn the bonus? (Do not count this
qualification HIT.)

7. How much will you earn in total by fully participating in this study, including the bonus? (Do not include earnings
from this qualification HIT.)

8. Each day's HIT will definitely be available during what times?

Specify a time range using Pacific Time:

Beginning at: Ending at:

Demographic survey
Your responses will kept anonymous and will not affect your participation in this study.

1. What is your gender?

Male Female Other

2. Would you say that in general your health is—

3. During the past month, other than your regular job, did you participate in any physical activities or exercises such
as running, calisthenics, golf, gardening, or walking for exercise?

4. On average, how many hours of sleep do you get in a 24-hour period?

5. What is your age?

6. What is the highest grade or year of school you completed?

Figure 5: Qualification session instructions (5 of 6)

10 Armed Forces, either in the regular military or in a

7. Have you ever served on active duty in the United States
National Guard or military reserve unit?

8. Are you currently…

 Male Female Other

2. Would you say that in general your health is—

3. During the past month, other than your regular job, did you participate in any physical activities or exercises such
as running, calisthenics, golf, gardening, or walking for exercise?

4. On average, how many hours of sleep do you get in a 24-hour period?

5. What is your age?

6. What is the highest grade or year of school you completed?

7. Have you ever served on active duty in the United States Armed Forces, either in the regular military or in a
National Guard or military reserve unit?

8. Are you currently…

Submit

Figure 6: Qualification session instructions (6 of 6)

11
Start today's HIT
This is the first of the 3 HITs for an academic research project.
Please review these instructions.

Earnings

This HIT pays $1.50

1 HIT this Wednesday; it pays $1.50
1 HIT next Wednesday; it pays $1.50
Earn a $5 bonus for completing all 3 HITs

This HIT's requirements

This HIT should only take 5-10 minutes to complete; it pays $1.50

1. Work 10 rows of the counting task (like the example below)

2. Decide how to split a workload of 360 rows of counting between Wednesday and next Wednesday

This Wednesday's HIT requirements

This HIT may take 10-20 minutes, depending on how you split the work between the days; it pays $1.50

1. Work 10 rows of the counting task (like the example below)

2. Decide how to split a workload of 360 rows of counting between Wednesday (that same day) and next
Wednesday
3. One of your decisions from today or Wednesday is selected to actually split the workload
4. Work the counting task for the selected amount of work for this Wednesday

Next Wednesday's HIT requirements

This HIT may take 10-20 minutes, depending on how you split the work between the days; it pays $1.50

1. Work 10 rows of the counting task (like the example below)

2. Work the counting task for the selected amount of work for next Wednesday

Summary

Earn $9.50 total (in addition to the $1.50 qual HIT) for about 30 to 40 minutes of work total, assuming you complete
all 3 HITs.

Example: working 10 rows of the counting task

Figure 7: Day zero session instructions (1 of 3)
This is a completed example of the counting task that would be in each HIT.

Please count the number of zeros (“0”) on each line and enter it in the box.

Each row will be marked correct or incorrect. You must correct errors before submission.

12
Row No. String Count (“0”)

1 8

2 12

3 8

4 7
 2. Work the counting task for the selected amount of work for next Wednesday

Summary

Earn $9.50 total (in addition to the $1.50 qual HIT) for about 30 to 40 minutes of work total, assuming you complete
all 3 HITs.

Example: working 10 rows of the counting task

This is a completed example of the counting task that would be in each HIT.

Please count the number of zeros (“0”) on each line and enter it in the box.

Each row will be marked correct or incorrect. You must correct errors before submission.

Row No. String Count (“0”)

1 8

2 12

3 8

4 7

5 10

6 8

7 10

8 2

9 5

10 8

Check responses and save

Decide how to split a workload between this Wednesday and next

Wednesday
You start with a workload of 360 rows of counting. You will decide how to split up the workload between this
Wednesday and Wednesday of next week.

You decide for different trade-off scenarios. For example:

Working 1 more row next week reduces work by 1 row this week (a 1-to-1 trade-off)
Working 1 more row next week reduces work by 1.5 rows this week (a 1-to-1.5 trade-off)
Working 1 more row next week reduces work by 0.5 rows this week (a 1-to-0.5 trade-off)

One scenario will be selected to actually matter.

You will get more details and practice in this HIT before you make your decisions that matter.

Complete each of Figure

3 days' HITs
8: Day zerofor a $5instructions
session bonus, (2 earning
of 3) $9.50 total
On each day in blue below, complete one HIT for 13$1.50.
Once you complete all 3 HITs, you will be paid a $5 bonus.
If you fail to complete a day's HIT, you will not be able to complete further HITs, nor will you receive the bonus.
You will always be paid for HITs you have already completed. All earnings are paid within 24 hours.
Each day's HIT will be available at least between 12 p.m. and 12 a.m. (midnight), Pacific Time.
You will receive a reminder notification through mTurk on each day that you have a HIT to complete.
Please do not accept more than one HIT on the same day; it will not give you additional work or earnings.
Please only participate if you think you can complete one HIT on each of these 3 days!

Sun, Oct 27 Mon, Oct 28 (today) Tue, Oct 29 Wed, Oct 30 Thu, Oct 31 Fri, Nov 1 Sat, Nov 2
Wednesday and Wednesday of next week.

You decide for different trade-off scenarios. For example:

Working 1 more row next week reduces work by 1 row this week (a 1-to-1 trade-off)
Working 1 more row next week reduces work by 1.5 rows this week (a 1-to-1.5 trade-off)
Working 1 more row next week reduces work by 0.5 rows this week (a 1-to-0.5 trade-off)

One scenario will be selected to actually matter.

You will get more details and practice in this HIT before you make your decisions that matter.

Complete each of 3 days' HITs for a $5 bonus, earning $9.50 total

On each day in blue below, complete one HIT for $1.50.
Once you complete all 3 HITs, you will be paid a $5 bonus.
If you fail to complete a day's HIT, you will not be able to complete further HITs, nor will you receive the bonus.
You will always be paid for HITs you have already completed. All earnings are paid within 24 hours.
Each day's HIT will be available at least between 12 p.m. and 12 a.m. (midnight), Pacific Time.
You will receive a reminder notification through mTurk on each day that you have a HIT to complete.
Please do not accept more than one HIT on the same day; it will not give you additional work or earnings.
Please only participate if you think you can complete one HIT on each of these 3 days!

Sun, Oct 27 Mon, Oct 28 (today) Tue, Oct 29 Wed, Oct 30 Thu, Oct 31 Fri, Nov 1 Sat, Nov 2
1 HIT 1 HIT

Sun, Nov 3 Mon, Nov 4 Tue, Nov 5 Wed, Nov 6 Thu, Nov 7 Fri, Nov 8 Sat, Nov 9
1 HIT

No deception is used in this experiment

This is an academic research project in the field of Economics, which widely prohibits the deception of experimental
subjects.

All information and instructions provided to you in this experiment are truthful.

If you feel anything in this experiment is deceptive, please notify the [IRB] at [IRB email].

If you have any questions about this research project, please contact [researcher] at [researcher email].

Begin practice round

You will complete a practice round before the actual tasks and decisions that matter.

Begin PRACTICE

Figure 9: Day zero session instructions (3 of 3)

14
PRACTICE MODE The correct answers are already filled in to save you time.

Complete 10 required rows of counting

Please count the number of zeros (“0”) on each line and enter it in the box.

Each row will be marked correct or incorrect. You must correct errors before submission.

Row No. String Count (“0”)

1 8

2 12

3 8

4 7

5 10

6 8

7 10

8 8

9 5

10 8

Check responses and save

Figure 10: Required tasks: This interface shows the subject examples of the task.

15
 PRACTICE MODE You will not have to work these tasks.

Split workload between Wed, Oct 30 and Wed, Nov 6

Choose how you want to split your workload of 360 rows of counting (in addition to the required 10 rows per
workday).

In this scenario, working 1 more row next week reduces work by 0.75 row(s) this week.

You're making five decisions on how to split the workload for Wed, Oct 30. You'll make five more similar decisions
on that day.

A coin flip will determine whether a decision made today or a decision made on Wed, Oct 30 will be selected to
actually matter.

One of today's five decisions may be randomly selected to actually split your workload.

The odds of this decision being the decision-that-matters are 10%.

Wed, Oct 30 Drag slider handle to adjust choice. Wed, Nov 6

256 rows 139 rows

Try moving the slider around to see how this trade-off rate splits your workload.

If this choice were selected to actually matter, your work schedule would be:

Sun, Oct 27 Mon, Oct 28 (today) Tue, Oct 29 Wed, Oct 30 Thu, Oct 31 Fri, Nov 1 Sat, Nov 2
10 rows required
+ 256 rows chosen
Sun, Nov 3 Mon, Nov 4 Tue, Nov 5 Wed, Nov 6 Thu, Nov 7 Fri, Nov 8 Sat, Nov 9
10 rows required
+ 139 rows chosen

You will be able to adjust this decision before finalizing it.

Continue

Figure 11: Task allocation, separate: This interface allows the subject to allocate their workload
between days.

16
 PRACTICE MODE You will not have to work these tasks.

Split workload between Wed, Oct 30 and Wed, Nov 6

You're making five decisions on how to split the workload for Wed, Oct 30. You'll make five more similar decisions
on that day.

A coin flip will determine whether a decision made today or a decision made on Wed, Oct 30 will be selected to
actually matter.

One of today's five decisions may be randomly selected to actually split your workload.

The odds of each decision being the decision-that-matters are 10%.

Trade-off Wed, Oct 30 Wed, Nov 6

1 to 0.5 360 rows 0 rows

1 to 0.75 274 rows 115 rows

1 to 1 139 rows 221 rows

1 to 1.25 40 rows 256 rows

1 to 1.5 0 rows 240 rows

Please review your choices and make any final changes.

Finalize

Figure 12: Task allocation, juxtaposed: This interface allows the subject to allocate their workload
between days.

17
 PRACTICE MODE

How today's decisions are used

You made decisions about splitting work between this Wednesday and next Wednesday.

You will make similar decisions again Wednesday. One day will be selected for its decisions to actually matter.

Sun, Oct 27 Mon, Oct 28 (today) Tue, Oct 29 Wed, Oct 30 Thu, Oct 31 Fri, Nov 1 Sat, Nov 2

Decisions made Decisions made

Sun, Nov 3 Mon, Nov 4 Tue, Nov 5 Wed, Nov 6 Thu, Nov 7 Fri, Nov 8 Sat, Nov 9

You just made five decisions about how to split work between these
days
Choice No. Trade-off Wed, Oct 30 Wed, Nov 6
1 1 to 0.5 360 rows 0 rows
2
3 Today's choices
1 to 0.75
1 to 1
235 rows
139 rows
167 rows
221 rows
4 1 to 1.25 52 rows 247 rows
5 1 to 1.5 0 rows 240 rows

You will make five similar decisions Wednesday

Choice No. Trade-off Wed, Oct 30 Wed, Nov 6
1 1 to 0.5 x rows x rows

Wednesday's choices
2 1 to 0.75 x rows x rows
3 1 to 1 x rows x rows
4 1 to 1.25 x rows x rows
5 1 to 1.5 x rows x rows

After you make decisions Wednesday, a coin-toss will select which

day's decisions are used
Choice No. Trade-off Wed, Oct 30 Wed, Nov 6 Choice No. Trade-off Wed, Oct 30 Wed, Nov 6
1 1 to 0.5 360 rows 0 rows 1 1 to 0.5 x rows x rows

Today's choices Wednesday's choices

2 1 to 0.75 235 rows 167 rows 2 1 to 0.75 x rows x rows
3 1 to 1 139 rows 221 rows 3 1 to 1 x rows x rows
4 1 to 1.25 52 rows 247 rows 4 1 to 1.25 x rows x rows
5 1 to 1.5 0 rows 240 rows 5 1 to 1.5 x rows x rows

Reveal

Figure 13: This interface gives the subject intuition regarding the selection procedure between days.

18
 PRACTICE MODE You will not have to work these tasks.

Implement workload for Wed, Oct 30 and Wed, Nov 6

For this practice round, the coin-toss selected today's decisions to actually matter.

Accordingly, here are the decisions you made today to split the workload of 360 rows of counting.

One rate and your corresponding choice is randomly selected to actually matter.

These work amounts are in addition to the 10 rows of counting required on each day.

Choice No. Trade-off Wed, Oct 30 Wed, Nov 6

1 1 to 0.5 360 rows 0 rows

2 1 to 0.75 235 rows 167 rows

3 1 to 1 139 rows 221 rows

4 1 to 1.25 52 rows 247 rows

5 1 to 1.5 0 rows 240 rows

Reveal

Figure 14: This interface gives the subject intuition regarding the selection procedure between rates.

19
PRACTICE MODE On Wednesday, you'd be asked to complete 235 tasks, if this wasn't practice.

On Wednesday, you would complete these counting tasks

Completed: 10 of 245 rows today (4%). Remaining: 235 tasks today.

Please count the number of zeros (“0”) on each line and enter it in the box.

Each row will be marked correct or incorrect. You must correct errors before submission.

Row No. String Count (“0”)

11 8

12 12

13 8

14 7

15 10

16 8

17 10

18 8

19 5

20 8

Check responses and save

Figure 15: This interface shows the subject examples of the implemented tasks.

20