Intuitive Generosity and Error Prone Inference from Response Time - - PowerPoint PPT Presentation

Intuitive Generosity and Error Prone Inference from Response Time

María P. Recalde1 Arno Riedl2 Lise Vesterlund1

1University of Pittsburgh 2Maastricht University

October 18, 2013

Motivation

When called upon to help others, is our instinctive, fast, gut reaction to help or to be selfish?

◮ Are we as individuals tempted to do good?

Existing work is consistent with temptation motives for giving

◮ DellaVigna, List, Malmendier (2012) ◮ Andreoni, Rao, Trachtman (2012)

Unclear how we can identify the intuitive response

◮ Rand, Greene, Nowak (2012) use decision time

Rand, Greene, and Nowak (2012)

4-person linear VCM, $0.40 endowment, mpcr=0.5, AMT Replicated by Lotito et al (2012) and Nielsen et al (2012)

The linear VCM

Can confound error with other-regarding preferences

◮ Location of the dominant strategy Nash equilibrium ◮ Location of efficiency maximizing strategy

Problematic if error correlates with response time

◮ Kocher & Sutter (2006) ◮ Rubinstein (2007), Agranov, Caplin, Tergiman (2012)

Our study

Research questions:

• 1. Is the correlation between generosity and decision time robust

to changes in the strategic environment?

• 2. Could decision error be playing a role?

Experiment:

◮ 4 person piece-wise linear VCM

unique interior Nash equilibrium in dominant strategies unique interior efficiency maximizing outcome

◮ between-subject design with two treatments

vary the location of the interior Nash equilibrium

Experimental design

Treatment Equilibrium contribution Low $3.00 out of $10.00 High $7.00 out of $10.00

Constant in both treatments:

◮ Individual endowment ($10.00), group size (4) ◮ Efficiency maximizing contribution ($9.00) ◮ Equilibrium payoffs ◮ Payoffs of contributing $0.00 and $10.00 ◮ Unit cost of deviation between giving $3.00 and $7.00

Implementation

◮ Neutral instructions without payoff information ◮ Tutorial on how to read a payoff table ◮ Actual payoff table presented on decision screen ◮ Response time measured since decision screen is displayed ◮ Two parts in a session

Part 1: one-time contribution decision Part 2: 10 repeated decisions with random rematching Same payoff table, one part randomly chosen for payment

Results - Part 1

Histogram of contribution decisions by treatment

.1 .2 .3 .4 share per treatment 1 2 3 4 5 6 7 8 9 10 contribution

Low High

Do we replicate Rand, Greene, and Nowak’s findings?

OLS regression of contribution on response time

Low response time

• 0.019**

(0.016) Constant 6.024*** (0.000) N 80

Note: P-values in parentheses. * p<0.10, ** p<0.05, *** p<0.01.

• Dep. Var.: Contribution to group account

Scatter plot of contribution and response time

50 100 150 200 250 300 2 4 6 8 10

Low

response time (seconds) contirbution

50 100 150 200 250 300

Does the correlation change with treatment?

OLS regression of contribution on response time and treatment

Low High All response time

• 0.019**

0.016**

• 0.019***

(0.016) (0.013) (0.010) High

• 0.205

(0.732) High x response time 0.035*** (0.001) Constant 6.024*** 5.819*** 6.024*** (0.000) (0.000) (0.000) Total effect decision time: High 0.016** (0.018) N 80 80 160

Note: P-values in parentheses. * p<0.10, ** p<0.05, *** p<0.01.

• Dep. Var.: Contribution to group account

Treatments

determinants of response time

Scatter plot of contribution and response time

50 100 150 200 250 300 2 4 6 8 10

Low

response time (seconds) contirbution

50 100 150 200 250 300 2 4 6 8 10

High

contribution

Contributions made by fast and slow decision makers

Contributions made by fast and slow decision makers

Contributions made by fast and slow decision makers

.1 .2 .3 share of treatment

Low High

Mistakes

fast slow

.1 .2 .3 share of treatment

Low High

Nash equilibrium

fast slow

.1 .2 .3 share of treatment

Low High

Midpoint of strategy space

fast slow

.1 .2 .3 share of treatment

Low High

Efficiency maximizing contribution

fast slow

Results - Part 2

Frequency of equilibrium play by period and treatment

.2 .4 .6 .8

percent frequency

1 2 3 4 5 6 7 8 9 10

period

Low High

Contributions - Part 2

Mean and median contribution by period and treatment

2 3 4 5 6 7 8 contribution 1 2 3 4 5 6 7 8 9 10 period

mean, Low C.I. mean, Low median, Low mean, High C.I. mean, High median, High

Response time - Part 2

Median response time by period and treatment

10 20 30 40 response time (seconds) 1 2 3 4 5 6 7 8 9 10 period

Low High

Correlation between contribution and response time

OLS regression of contribution on response time

decision time 0.014 0.022 0.014 (0.340) (0.278) (0.261) period

• 0.068**

0.047*

• 0.068***

(0.047) (0.052) (0.010) High NE 2.143*** (0.003) High NE X decision time 0.009 (0.668) High NE X period 0.115*** (0.002) Constant 4.082*** 6.225*** 4.082*** (0.001) (0.001) (0.000) Total effect decision time: High NE 0.022 (0.196) Total effect period: High NE 0.047** (0.012) R-squared 0.041 0.025 0.506 N 800 800 1600 Low NE High NE All Treatment

determinants of decision time

Contributions made by fast and slow decision makers

Mean contribution by response time in part 1

3 4 5 6 7 contribution

1 2 3 4 5 6 7 8 9 10

period fast in part 1, Low slow in part 1, Low fast in part 1, High slow in part 1, High

Response time of fast and slow decision makers

Median response time

20 40 60

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Low High

fast in part 1 slow in part 1 fast in part 1 slow in part 1

response time (seconds)

period

Conclusion

◮ The correlation between choices and response time changes

with features of the decision environment

◮ Error negatively correlates with response time ◮ Potential explanation for mixed evidence in the literature

Piovesan and Wengstrom (2008), Fiedler et al (2013) Tinghog et al (2013), Rand et al (2013) Matthey and Regner (2010)

◮ Caution warranted when making inferences about preferences

from response time

Thank you

Payoff table - Low treatment

resume

Payoff table - High treatment

resume

Screen shot - Tutorial

resume

Determinants of response time, part 1

Low High All % tutorial correct 22.121 30.526 22.121 (0.328) (0.165) (0.331) experiments

• 0.824
• 0.948
• 0.824

(0.125) (0.148) (0.126) econ courses 1.175 5.763*** 1.175 (0.396) (0.004) (0.398) High

• 25.062

(0.779) High x % tutorial correct 8.404 (0.789) High x experiments

• 0.123

(0.883) High x econ courses 4.588* (0.055) Constant

• 2.218
• 27.280
• 2.218

(0.972) (0.667) (0.972) Age and gender controls Yes Yes Yes N 80 80 160

resume

Determinants of response time, part 2

Low High All period

• 1.564***
• 1.249**
• 1.573***

(0.004) (0.021) (0.000) % past equilibrium play 1.422 3.377* 1.835** (0.213) (0.075) (0.020) mean contribution others

• 0.130

0.353

• 0.135

(0.843) (0.215) (0.816) experiments

• 0.268**
• 0.136**
• 0.274***

(0.031) (0.045) (0.005) High

• 6.528

(0.599) High x period 0.322 (0.338) High x % past equilibrium play 1.142 (0.330) High x mean contribution others 0.484 (0.439) High x experiments 0.142 (0.114) Constant 15.762 9.532 15.639 (0.178) (0.394) (0.100) Age and gender controls Yes Yes Yes Econ training and tutorial controls Yes Yes Yes N 800 800 1600

resume