Econ 219B Psychology and Economics: Applications (Lecture 3) · 1/31/2018 · Randomization of pay...

Econ 219B

Psychology and Economics: Applications

(Lecture 3)

Stefano DellaVigna

January 31, 2018

Stefano DellaVigna Econ 219B : Applications (Lecture 3) January 31, 2018 1 / 100

Outline

1 Investment Goods: Exercise II

2 Investment Goods: Job Search

3 Investment Goods: Work Effort

4 Investment Goods: Delay with Deadline

5 Leisure Goods: Credit Card Borrowing

6 Leisure Goods: Consumption and Savings

7 Leisure Goods: Commitment and Savings

8 Leisure Goods: Drinking

9 Methodology: Commitment Field Experiments


Investment Goods: Exercise II

Section 1




Explanation

Present-Biased preferences with naivete explains magnitudes,not just qualitative patterns

Related: Acland and Levy (MS 2015) field experimentPay a treatment group $100 to attend the gym for 4 weeksControl group not paid


Investment Goods: Exercise II Acland and Levy 2015

Results

Moderate habit formation (as in Charness and Gneezy EMA)



Expectation of Future Attendance

Also: Elicit expectation of future attendance as WTP forp-coupons

How much are you willing to pay for coupon below?Also elicit unincentivized forecasts



Clear Evidence of Naivete

Evidence of naivete consistent with DVMAlso some evidence of projection bias (see later lectures)


Investment Goods: Job Search

Section 2

Investment Goods: Job Search


Investment Goods: Job Search DellaVigna and Paserman 2005

DellaVigna and Paserman (JOLE 2005)

Stylized facts:time devoted to job search by unemployed workers: 9hours/week

search effort predicts exit rates from unemployment better thanreservation wage choice

Model with costly search effort and reservation wage decision:

search effort — immediate cost, benefits in near future — drivenby βreservation wage — long-term payoffs — driven by δ



Correlations

Correlation between measures of impatience (smoking,impatience in interview, vocational clubs) and job searchoutcomes:

Impatience ↑ =⇒ search effort ↓Impatience ↑ =⇒ reservation wage ←→Impatience ↑ =⇒ exit rate from unemployment ↓

Impatience captures variation in βSophisticated or naive – does not matter



Correlations


Investment Goods: Job Search Paserman 2008

Paserman (EJ 2008)

Structural model estimated by max. likelihoodEstimation exploits non-stationarity of exit rate fromunemployment


Investment Goods: Work Effort

Section 3

Investment Goods: Work Effort


Investment Goods: Work Effort Kaur, Kremer, and Mullainathan (2015)

”Self-Control at Work”

Kaur, Kremer, and Mullainathan, “Self Control at Work.” JPE 2015

Setting: workers in India who are paid a piece rate w in a weeklypaycheck

Since effort at work is immediate and benefits delayed, effort atwork is an investment good(β, β, δ

)model, with δ = 1

Consider effort at work e, which costs −c (e) , with c ′ > 0,c ′′ > 0

Assume for special case c (e) = γe2/2



Model

Two states:

high output yH with probability e pay wH

low output yL with probability 1− e pay wL

Notice: this is only local approximation, for e ∈ [0, 1]

Pay at t = 2

If working at t = 1, maximize

maxe1

β [e1wH + (1− e1)wL]− c (e1)

f.o.c.β [wH − wL]− c ′ (e∗1 ) = 0

Effort e∗1 increases in wH − wL and in β



Model

Special case:

e∗1 =β [wH − wL]

γ

If working at t = 2 (same period as paydate), optimal effort e∗2 solves

maxe2

[e2wH + (1− e2)wL]− c (e2)

and thus (for the special case)

e∗2 =wH − wL

γ

Prediction 1. Effort is higher near payday for β < 1 (independent ofβ)

From t = 0 perspective, (perceived) utility V0 from working at t = 1is

V0 = e∗1wH + (1− e∗1 )wL − c (e∗1 )



Model

Effect of altering wL on t = 0 (expected) welfare V0 is

dV0

dwL= (1− e∗1 ) +

de∗1dwL

[[wH − wL]− c ′ (e∗1 )] =

= (1− e∗1 ) +de∗1dwL

[(1− β

)[wH − wL]

]First term is direct effect on pay: lowering wL lowers pay andthus welfareThe second term is the effect on incentive, which is zero forβ = β = 1, by the envelope theorem – but envelope theoremdoes not apply for β < 1. Indeed, second term is negativeNotice that it is β which matters, since this is the valuefunction from the t = 0 perspective



Model

Special case for β = β:

dV0

dwL= 1− β [wH − wL]

γ− β (1− β) [wH − wL]

γ

Second term becomes large as β goes below 1 and is highest atβ = 1/2If large enough, individual wants commitment device, prefers wL

low

Prediction 2. Individual with β < 1 may prefer commitmentdevice (low wL)

Prediction 3. If there are both types with β = 1 and β < 1,demand for commitment should be associated with a paydaycycle



Experiment

Field experiment in India

Randomization of pay date (Tu, Th, Sa) to test proposition 1unconfounded with day-of-week effectsRandomization of availability of commitment device: get paidw/2 instead of w if miss production targetRandomization of whether choice is made evening before, ormorning of



Predictions

Prediction 1. Evidence of pay cycle in effort



Predictions

Prediction 2. Quite significant take-up of commitment contract



Results



Predictions

Prediction 3. Correlation between payday effect and take-up ofcommitment, as well as with productivity effect



Results



Summary

Evidence very consistent with model of self-control problems and(at least partial) sophistication

Discount factor is not β − δ, but smoother decay (truehyperbolic)

Significant demand of commitment device – different than someof other settings, see later

Correlation with underlying measure of self-control

Great evidence in important setting


Investment Goods: Delay with Deadline

Section 4




Setting

Consider as individual that has to do an unpleasant task, with afixed deadline

Examples:

Paying a traffic fine (Heffetz, O’Donoghue, and Schneider,2016)Filing taxes (Martinez, Meier, and Sprenger, 2017; Benzarti,2016)Finding a job by the deadline (job search papers)

What can we infer from spike at deadline?


Investment Goods: Delay with Deadline Heffetz, O’Donoghue, and Schneider (2016)

Paying a Traffic fine with multiple deadlines0

.01

.02

.03

.04

.05

Haza

rd ra

te

0 10 20 30 40 50 60 70 80 90 100 110 120 130Days since issue date

0.2

.4.6

.81

Cum

ulat

ive re

spon

se ra

te

0 10 20 30 40 50 60 70 80 90 100 110 120 130Days since issue date

OLD regime NEW regime

Figure 1: Hazard Rates and Cumulative Response Rates in OLD vs. NEWRegimes. Note: All tickets have a first deadline at day 30, second deadline at days 62-68, andthird deadline at days 101-107, indicated by the shaded areas (the latter two deadlines are a rangebecause they depend on ticket-issuance day of the week). First notification letter is received aroundday 40 (OLD) vs. day 20 (NEW). Based on 3,355,094 (OLD) and 3,020,357 (NEW) observations;see details in Section 3.

While the pattern in Figure 1 is consistent with forgetting, the specific way in which DOF

implemented the regime change resulted in issues of interpretation. To rule out alternative

explanations, we worked with DOF on a field experiment that was implemented over five

weeks, for tickets issued July 13, 2013 through August 16, 2013. The recipient of each ticket

was assigned to receive one of four versions of the first letter (at roughly day 20), and to

receive or not receive an additional letter at roughly day 48 (a 4 × 2 experimental design).

The four versions of the first letter vary in their informational content and presentation. The

results seem to confirm that the letters served primarily as reminders.

Figure 1 exhibits other important patterns. First, under both regimes, there is a dramatic

decline in hazard rates between the first and second deadlines. While this decline could be

due in part to forgetting, a similar decline in hazard rates over time is typically observed in

job search data, and has been attributed to heterogeneity in types (Salant (1977)). Second,

while there is a noticeable spike in hazard rates at the first deadline under both regimes

(and especially under the NEW regime), there are at most small spikes in hazard rates at

the second and third deadlines. This apparent limited response to later deadlines is also

3

Figure 1: Hazard Rates and Cumulative Response Rates in OLD vs. NEW Regimes. Note: All tickets have a first deadline atday 30, second deadline at days 62-68, and third deadline at days 101-107, indicated by the shaded areas (the latter twodeadlines are a range because they depend on ticket-issuance day of the week). First notification letter is received around day40 (OLD) vs. day 20 (NEW). Based on 3,355,094 (OLD) and 3,020,357 (NEW) observations; see details in Section 3.


Investment Goods: Delay with Deadline Martinez, Meier, and Sprenger (2017)

Filing Taxes by April 15

500

1500

2500

500

1500

2500

01

23

01

23

0 20 40 60 80 0 20 40 60 80

2005 2006

2007 2008

Percentage of Filers Average Refund Value

Perc

enta

ge

Day of Season

Notes: 2005-2008 percentage of filers on each day of tax season (gray bars) and average refund value for filers on each

day (black line).

Figure 1: Filing Times and Refund Values

26

Page 30 of 85


Investment Goods: Delay with Deadline Martinez, Meier, and Sprenger (2017)

Inference

What can we learn from extent of last-minute completion?

If there is a lot of last-minute completion, does that indicateprocrastination?

Or could it indicate a particular distribution of the cost of doingthe action?

Heffetz, O’Donoghue, and Schneider, 2016:

Use multiple traffic finesThere are types that delay more on multiple traffic ticketinfractions

Martinez, Meier, and Sprenger (2017)

Structural estimation of time preference parameters using alsovariation in amount of tax refund and delay in refund


Leisure Goods: Credit Card Borrowing

Section 5

Leisure Goods: Credit Card Borrowing


Leisure Goods: Credit Card Borrowing Ausubel (1999)

Ausubel, “Adverse Selection in Credit Card

Market”

Joint-venture: company-researcher

Field Experiment: Randomized mailing of two millionsolicitations!

Follow borrowing behavior for 21 months

Variation of:

pre-teaser interest rate r0: 4.9% to 7.9%post-teaser interest rate r1: Standard - 4% to Standard +4%Duration of teaser period Ts (measured in years)



Design

Part of the randomization – Incredible sample sizes. How muchwould this cost to run? Millions



Design

Another set of experiments:



Model

Setting:

Individual has initial credit card (r00 , r

01 ,T

0s ). Balances: b0

pre-teaser, b1 post-teaserCredit card offers: (r ′0, r

′1,T

′s)

Decision to take-up new credit card:

switching cost k > 0approx. saving in pre-teaser rates (Ts years): Ts

(r ′0 − r0

0

)b0

approx. saving in post-teaser rates (21/12− Ts years):(21/12− Ts) (r

′1 − r1)b1

Net benefit of switching:

NB ′ = −k + Ts

(r′

0 − r 00

)b0 + (21/12− Ts)

(r′

1 − r 01

)b1



Model

Switch if NB + ε > 0

Take-up rate R is function of attractiveness NB :

R = R (NB) , R ′ > 0

Compare take-up rate of card i , R i , to take-up rate of StandardCard St, RSt

Standard Card (6.9% followed by 16%) (Card C above)

Assume R (approximately) linear in a neighborhood of NBSt ,that is,

R(NB i

)= R

(NBSt

)+ R ′NB

(NB i − NBSt

)Stefano DellaVigna Econ 219B : Applications (Lecture 3) January 31, 2018 38 / 100


Model

Compare cards Pre and St that differ only in interest rate r0(pre-teaser)

Assume bPre0 = bSt0 = b0 (Pre-teaser balance ) ≈ $2, 000

Difference in attractiveness:

R(NBPre

)− R

(NBSt

)= R ′NBTs

(rPre0 − rSt0

)b0

Pre-Teaser Offer (Card A): (4.9% followed by 16%)

NBPre − NBSt ≈ 6/12 ∗ 2% ∗ $2, 000 = $20R(NBPre

)− R

(NBSt

)= 386 out of 100,000



Model

Compare cards Post and St that differ only in interest rate r1(post-teaser)

Assume bPost1 = bSt1 = b1 (Post-teaser balance) ≈ $1, 000

Difference in attractiveness:

R(NBPost)− R(NBSt) = R ′NB (21/12− Ts)(rPost1 − rSt1

)b1

Post-Teaser Offer (Card B in Exp. III): (6.9% followed by 14%)

NBPost − NBSt ≈ 15/12 ∗ 2% ∗ $1000 = $25R(NBPost)− R(NBSt) = 154 out of 100,000

Puzzle:

NBPost − NBSt > NBPre − NBSt

But R(NBPre)− R(NBSt) >> R(NBPost)− R(NBSt)



Results

Plot NB and R(NB) for different offersCompare offers varying in r0 (flat line) and in r1 (steep line)



Results

People underrespond to post-teaser interest rate.

Most likely explanation: Present Bias + Naivete

Naives overestimate switching to another card (procrastination) Underestimate post-teaser borrowing: b1 < b1 and b0 = b0

Compare cards:

NBPre − NBSt = Ts

(rPre0 − rSt0

)b0

and

NBPost − NB

St= (21/12− Ts)

(rPost1 − rSt1

)b1

Calibration: b1 ≈ (1/3) b1 Underestimation of borrowing by afactor of 3


Leisure Goods: Consumption and Savings

Section 6




Introduction

Laibson (1997) to Laibson, Repetto, and Tobacman (2007)

Leisure Good: Temptation to overconsume at present

Stylized facts:

Low liquid wealth accumulationExtensive credit card borrowing (SCF, Fed, Gross and Souleles2000)Consumption-income excess comovement (Hall and Mishkin,1982)Substantial illiquid wealth (housing+401(k)s)



Reduced-form evidence here not sufficient

Life-cycle consumption model (Gourinchas and Parker, 2004)

Assume realistic features:

borrowing constraintsilliquid assetsbequests...

David Laibson’s slides to follow


3.1 Demographics

• Mortality, Retirement (PSID), Dependents (PSID),HS educational group

3.2 Income from transfers and wages

• Yt = after-tax labor and bequest income plus govttransfers (assumed exog., calibrated from PSID)

• yt ≡ ln(Yt). During working life:yt = f

W (t) + ut + νWt (3)

• During retirement:yt = f

R(t) + νRt (4)

3.3 Liquid assets and non-collateralized debt

• Xt + Yt represents liquid asset holdings at the

beginning of period t.

• Credit limit: Xt ≥ −λ · Yt

• λ = .30, so average credit limit is approximately

$8,000 (SCF).

3.4 Illiquid assets

• Zt represents illiquid asset holdings at age t.

• Z bounded below by zero.

• Z generates consumption flows each period of

γZ.

• Conceive of Z as having some of the properties

of home equity.

• Disallow withdrawals from Z; Z is perfectly

illiquid.

• Z stylized to preserve computational tractability.

3.5 Dynamics

• Let IXt and IZt represent net investment into as-

sets X and Z during period t

• Dynamic budget constraints:Xt+1 = RX · (Xt + IXt )Zt+1 = RZ · (Zt + IZt )Ct = Yt − IXt − IZt

• Interest rates:

RX =

(RCC if Xt + I

Xt < 0

R if Xt + IXt > 0

; RZ = 1

• Three assumptions forhRX, γ, RCC

i:

Benchmark: [1.0375, 0.05, 1.1175]Aggressive: [1.03, 0.06, 1.10]Very Aggressive: [1.02, 0.07, 1.09]

In full detail, self t has instantaneous payoff function

u(Ct, Zt, nt) = nt ·³Ct+γZtnt

´1−ρ − 11− ρ

and continuation payoffs given by:

βT+N−tXi=1

δi³Πi−1j=1st+j

´(st+i) · u(Ct+i, Zt+i, nt+i)...

+βT+N−tXi=1

δi³Πi−1j=1st+j

´(1− st+i) ·B(Xt+i, Zt+i)

• nt is effective household size: adults+(.4)(kids)

• γZt represents real after-tax net consumption flow

• st+1 is survival probability

• B(·) represents the payoff in the death state

3.7 Computation

• Dynamic problem:maxIXt ,I

Zt

u(Ct, Zt, nt) + βδEtVt,t+1(Λt+1)

s.t. Budget constraints

• Λt = (Xt + Yt, Zt, ut) (state variables)

• Functional Equation:Vt−1,t(Λt) =st[u(Ct, Zt, nt)+δEtVt,t+1(Λt+1)]+(1−st)EtB(Λt)

• Solve for eq strategies using backwards induction

• Simulate behavior

• Calculate descriptive moments of consumer be-havior

4 Estimation

Estimate parameter vector θ and evaluate models wrt data.

• me = N empirical moments, VCV matrix = Ω

• ms (θ) = analogous simulated moments

• q(θ) ≡ (ms (θ)−me)Ω−1 (ms (θ)−me)0, a scalar-valued loss function

• Minimize loss function: θ = argminθq(θ)

• θ is the MSM estimator.

• Pakes and Pollard (1989) prove asymptotic con-sistency and normality.

• Specification tests: q(θ) ∼ χ2(N−#parameters)


Two steps of estimation: of MSM (Method of SimulatedMoments)

1 Estimate (‘calibrate’) auxiliary parameters

Interest rateMortalityIncome shocks

2 Estimate main parameters (β, δ) using Method of SimulatedMoments

Simulate model (cannot solve analytically)Choose parameters (β, δ) that minimize distance of simulatedmoments to estimated momentsTake into account uncertainty in estimates of 1st stage


Leisure Goods: Commitment and Savings

Section 7

Leisure Goods: Commitment and Savings


Leisure Goods: Commitment and Savings Ashraf, Karlan, and Yin (2005)

Ashraf, Karlan, and Yin (2005), QJE

Different Methodology: Commitment Device Field Experiment

Different Setting: Philippines

Three treatments:

SEED Treatment (N=842): Encourage to save, Offercommitment device (account with savings goal)Marketing Treatment (N=466): Encourage to save, Offer nocommitmentControl Treatment (N=469)



Results

Result 1. Take-up of commitment device (in SEEDTreatment):

Out of 842 treated people, 202 take up SEED Take up of24%167 also got lock-up box (did not observe savings there)

Result 2. Effect of Availability of Commitment on TotalSavings (including funds in non-committed account)

Compare SEED to Marketing (Include all 842 people,Intent-to-Treat)Share of people with increased Balances: 5.6 percentage(33.3 percent in SEED and 27.7 in Marketing)Share of people with increased Balances by at least 20 percent:6.4 percentage pointsTotal Balances: 287 Pesos after 6 months (not significant)

To compute Treatment-on-The-Treated, divide by 202/842



Results



Results

Survey response to hyperbolic-discounting-type question:

Preference between 200 Pesos now and in 1 monthPreference between 200 Pesos in 6 months and in 7 monthsOn average, evidence of hyperbolic-discounting-type preferences



Results: Selection

Result 3. Who takes up the Commitment device?Correlate survey response with commitment take-up (see alsoFehr-Goette paper)

Evidence of correlation for women, not for men


Leisure Goods: Drinking

Section 8

Leisure Goods: Drinking


Leisure Goods: Drinking Schilbach (2015)

Schilbach (2015)

Consider population with high levels of drinking while working

Offer incentives to not drink during work hours

Examine impact on

Drinking during work hoursDrinking after work hoursEarningsSavings with (and without) savings commitment

Frank’s slides follow


Introduction Background Design Alcohol Impact Commitment Outlook

Heavily concentrated alcohol consumption in India

0.7

5.0

1.4

2.8

2.3

4.9

01

23

45

Sta

ndard

drinks p

er

day

India USA Russia

Source: WHO (2014), WHO (2001), own calculation.

Per capita (age 15+) Male drinkers only

6 / 53


Study sample

• Cycle-rickshaw peddlers in Chennai• 35 years old, 5 years of education• 80% are married, 2 children• Average daily labor incomes of about Rs. 300 ($5)

• Alcohol consumption• Individuals drink (almost) every day, usually alone.• A third of labor incomes spent on hard liquor (>80 proof)• Individuals drink over 5 standard drinks per day.• High levels of intoxication, often during the day• 80% say they would be better off if all liquor stores closed.

SELECTION TABLE

12 / 53


Experimental design

• 229 individuals paid to visit study office for 20 days• Daily visits any time between 6 pm and 10 pm• Measure blood-alcohol content (BAC) using breathalyzer test• Short survey

• Labor market outcomes• Alcohol consumption• Expenditure patterns

• Opportunity to save money at study office

13 / 53


Financial incentives for sobriety: three treatment groups

(I) Control Group: unconditional payments• Paid Rs. 90 regardless of BAC

(II) Incentive Group: monetary incentives to show up sober• Paid Rs. 60 if BAC > 0• Paid Rs. 120 if BAC = 0

(III) Choice Group• Choice between incentives and unconditional payments

BALANCE TABLES

14 / 53


Experimental design

-Day 1 Day 4 Day 7 Day 13 Day 20?

BaselineConsentScreening

?

Incentivesassigned

?

Choice 1

?

Choice 2

?

Choice 3Endline

Control

AAAAU

Incentives

(2/3)

Control

(1/3)

Incentives(1/3)J

JJJJ

Choice(1/3)

Control(1/3)

-

-

Choice-

CCCCW

-

Choice

15 / 53


Financial incentives significantly increased daytime sobriety.

← Alcohol treatment assigned

30

40

50

60

Fra

ction s

ober

(%)

0 5 10 15 20Day in Study

Incentives Choice Control

Sobriety at the Study Office

ATTENDANCE IN THE INCENTIVE GROUP IS LOWER.

18 / 53


...but reported overall drinking did not fall by much.

↑

drinking before study office visit

overall drinking

↓

01

23

45

67

Num

ber

of sta

ndard

drinks


Incentives and Choice Groups pooled Control Group

# Standard Drinks Before Study Office Visit and Overall

20 / 53

Appendix

Intertemporal substitution: time of first drink

0.2

.4.6

.81

cdf

6 8 10 12 14 16 18 20 22 24

Time of first drink

Incentives Choice Control

BACK TO SUMMARY OF EFFECTS ON DRINKING

69 / 53


No significant effects on earnings


250

270

290

310

330

Daily

earn

ings (

Rs.)



Earnings (Rs/day)

25 / 53


Measuring the impact of increased sobriety on savings

• All subjects got personalized savings box at study office.• Could save up to Rs. 200 per day.• Paid out entire amount plus matching contribution on day 20.

• Cross-randomized matching contribution to benchmark effects• 10% vs. 20% of amount saved

• Cross-randomized commitment savings feature• Allowed to withdraw any day between 6 pm and 10 pm• Not allowed to withdraw until day 20

29 / 53


Incentives for sobriety increased savings.


0200

400

600

Cum

ula

tive s

avin

gs a

t stu

dy o

ffic

e (

Rs)



Cumulative Savings by Treatment Group

DEPOSITS WITHDRAWALS

31 / 53


Incentives for sobriety increased savings.

(1) (2) (3)VARIABLES Rs/day Rs/day Rs/day

Pooled alcohol treatment 12.45** 13.41*** 11.55**(6.262) (5.018) (4.792)

High matching contribution 9.29 10.11** 11.65**(6.532) (4.873) (4.619)

Commitment savings 7.59 2.88 2.86(6.539) (5.074) (4.820)

Daily study payment (Rs) 0.35***(0.050)

Observations 3,435 3,435 3,435R-squared 0.006 0.113 0.129Baseline survey controls NO YES YESPhase 1 controls NO YES YESControl mean 20.42 20.42 20.42

Standard errors in parentheses, clustered by individual.32 / 53


Interaction between sobriety and commitment savings

0100

200

300

400

500

600

700

Cum

ula

tive s

avin

gs (

Rs)


Pooled alcohol treatment, commitment savings

Pooled alcohol treatment, no commitment savings

No alcohol treatment, commitment savings

No alcohol treatment, no commitment savings

Sobriety Incentives vs. Commitment Savings

WITHDRAWALS DEPOSITS

39 / 53

Appendix

Sobriety incentives vs. commitment savings: deposits

0200

400

600

800

Cum

ula

tive d

epsits (

Rs)


Sobriety incentives, commitment savings

Sobriety incentives, no commitment savings

No sobriety incentives, commitment savings

No sobriety incentives, no commitment savings

Sobriety vs. Commitment Savings: Cumulative Deposits

BACK TO SAVINGS SECTION

65 / 53

Appendix

Sobriety incentives vs. commitment savings: withdrawals

050

100

150

200

Cum

ula

tive w

ithdra

wals

(R

s)


Sobriety incentives, commitment savings

Sobriety incentives, no commitment savings

No sobriety incentives, commitment savings

No sobriety incentives, no commitment savings

Sobriety vs. Commitment Savings: Cumulative Withdrawals

BACK TO SAVINGS SECTION

66 / 53


Eliciting willingness to pay for incentives

• Choice Group chooses between:• Incentives for sobriety• Unconditional payments

• Choice sessions on days 7, 13, 20, each for subsequent week• Elicit preferences for set of 3 choices• Then randomly select one choice to be implemented (RLIS)

41 / 53


Demand for incentives

• Option A: incentives for sobriety• Same payment structure as Incentive Group• Rs. 60 if BAC > 0, Rs. 120 if BAC = 0

• Option B: payment of Rs. Y regardless of BAC

Option A Option BBAC > 0 BAC = 0 regardless of BAC

(1) Rs. 60 Rs. 120 Rs. 90(2) Rs. 60 Rs. 120 Rs. 120(3) Rs. 60 Rs. 120 Rs. 150

42 / 53


Demand for commitment persists over time.

0.2

.4.6

.8F

raction o

f C

hoic

e G

roup w

ho c

hose incentives

1 2 3 1 2 3 1 2 3Week

Choice 1: unconditional payment = Rs 90



Demand for Incentives over Time

46 / 53


Exposure to incentives increases demand for incentives.

0.2

.4.6

.8F

raction o

f in

div

iduals

who c

hose incentives

Choice 1 (Rs 90) Choice 2 (Rs 120) Choice 3 (Rs 150)

Incentive Group Choice Group Control Group

Demand for Incentive across Treatment Groups

50 / 53

Leisure Goods: Drinking Schilbach (2015)

Unique Features

Unique feature 1: Effect of commitment device on drinking onanother patience-related activity: savings

How do we interpret the effect?

Effect on withdrawing – mechanical given drunkennessEffect on deposits – sophistication?

Would be great to know if sobriety incentives increases or lowersdemand for savings commitment device (not in design)

Unique feature 2: Exceptional demand for commitment device byfor drinking

1/3 population even when very expensiveOther existing studies – Demand typically goes to near zero


Methodology: Commitment Field Experiments

Section 9




Growing literature on field experiments offering

commitment devices

Recipe for typical device:

Random assignment into Treatment (T) and Control (C)Group T: Offered commitment option (action that imposesconstraints)Group C: No optionObserve take-up of commitment in TObserve outcome (e.g., saving, smoking, eating) in C and T



Three sets of results

1 Take-up. What share in T uses commitment device?Standard agent would not choose additional constraints Smoking gun for time inconsistencyTime inconsistency can be from present bias+ sophisticationOR from hot/cold states or intra-family bargaining

2 Effect on outcome. Compare outcomes in T and CNotice: Compare everybody in T to everybody in CCannot focus on those that took up the commitment in T, sincedo not know who they compare to in CTreatment on Treated: rescale by dividing by take-up(assumption of no effect on non-takers)

3 Who Takes Up? Document who in T takes up commitmentCorrelation with measured time preferences, previous behavior,etc.This is not causal evidence, but still interesting



Representative studies: Investment Goods

Homework Completion (Ariely-Wertenbroch PS)

Deadlines are penalties for delivering homework lateResult 1. Very large take-up rate (65 percent)Result 2. Large effect on quality of homework and delay (in exp.2)



Representative studies: Investment Goods

Health-club attendance (Royer, Stehr, and Sydnor, AEJ Applied2014)

First pay a treatment group to go to the gymThen offer half of this treated group commitment device tokeep goingCommitment device is money deposited into an account.Money forfeited if do not attend at least once every 14 days for4 monthsResult 1: Low demand for commitment: 13% take-up, withaverage sum of $63Result 2: Some effect on attendance



Health-club attendance0

.51

1.5

22.5

3

-20 -10 0 10 20Week Relative to Start of Incentives

Control IncentiveIncentive+Commit

Average Weekly Visits Overall



Representative studies: Leisure Goods

Consumption/Savings (Ashraf-Karlan-Yin)

Result 1. Commitment device take-up 24%Result 2. Significant effect on overall savings

Consumption/Savings (Beshears, Choi, Laibson, Madrian,Mekong, 2011)

RAND panel respondents, 495 subjects, given $50, $100, or$500Choice between

Liquid account (r=22% yearly)Commitment account (set a goal) with r of 21%, 22%, or 23%Penalty for early withdrawal(Notice: only group with r=21% is a commitment device design)Can choose share into each account



Beshears et al., 2011

Result 1. Commitment device take-up quite high – up to 56%




Retirement Savings (SMRT plan, Thaler and Benartzi, 2007 –last lecture)

Result 1. Take-up rate 80% when offered in personResult 2. Huge effects on 401(k) contribution rates

Online gaming (Chow, 2010 and Acland and Chow, 2010)

Offer online interface that one can use to limit play of onlinegamesResult 1. Take-up rate relatively high initially, but declines to5-10%Result 2. Suggestive effects on time spent playing




Smoking (Gine, Karlan, and Zinman, 2010)

Offer urine test for smoking in 6 monthsCan deposit money into account – forfeited if fail test at month6Result 1. Low take-up: 11% of 781 offered productResult 1. Conditional on take-up, average deposit of 57 pesos(4 weeks worth of cigarettes)Result 2: At 6 months, increase of 4-5 percentage point inchance of making urine test




Smoking (Gine, Karlan, and Zinman, 2010), continued

Result 2: At 12 months, similar increase at surprise test




Why often low-take up? At least 3 possibilities:

Self-control not prevalentSelf-control prevalent, but naivete’ is strongDemand for commitment outweighed by costs of commitmentin terms of loss of flexibility

Important to have designs to separate explanations




Alternative design of the commitment device field experiments:2*2 Design (Chow, 2010)

Offer everyone the commitment deviceThen randomly assign whether commitment device is actuallyofferedTherefore groups are 2 (wanted comm./did not) * (gotcomm./did not)

Advantage of this design

More power on demand for commitment since everybody (notjust 1/2 of subjects) is askedCan estimate effect of commitment both on the subjects thatdemand it, and the ones who do not (but who may end upusing it)See also Chassang, Padro-i-Miguel, Snowberg, (AER 2012)


Next Lecture

Section 10

Next Lecture


Next Lecture

Next Lecture

Present Bias

Laboratory ExperimentsErrors in Applying Present Bias

Then Reference-Dependent Preferences

HousingBunching-based Evidence


Econ 219B Psychology and Economics: Applications (Lecture 3) · 1/31/2018 · Randomization of pay...

Documents

Transcript of Econ 219B Psychology and Economics: Applications (Lecture 3) · 1/31/2018 · Randomization of pay...