Cognitive Load and Strategic Sophistication Sarah Allred Sean Duffy John Smith Psychology Psychology...
-
Upload
lexie-turville -
Category
Documents
-
view
213 -
download
1
Transcript of Cognitive Load and Strategic Sophistication Sarah Allred Sean Duffy John Smith Psychology Psychology...
Cognitive Load and
Strategic Sophistication
Sarah Allred Sean Duffy John SmithPsychology Psychology Economics
Rutgers University-Camden
Beauty Contest-Laboratory Outcomes
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95100
2
Models of Strategic Sophistication Level-k/Cognitive Hierarchy Models
Stahl and Wilson, 1994, 1995; Nagel, 1995; Costa-Gomes et al., 2001; Camerer et al., 2004
Different levels of strategic sophistication
Very successful in interpreting heterogeneous experimental data
3
Strategic Sophistication (continued) Does the observed strategic sophistication
relate to a fundamental characteristic of the subject?
Examine relationship between measures of cognitive ability and strategic behavior
Ballinger et al. (2011), Bayer and Renou (2011), Brañas-Garza et al. (2012), Brañas-Garza et al. (2011), Burnham et al. (2009), Carpenter et al. (2013), Chen et al. (2009), Chen et al. (2011), Devetag and Warglien (2003), Georganas et al. (2010), Gill and Prowse (2012), Jones (2011), Jones (2008), Palacios-Huerta (2003), Putterman et al. (2011) and Rydval (2011)
4
Experimental Rather than measure cognitive ability
We manipulate it
Advantage to manipulating cognitive ability Cognitive ability related to lots of other things Maybe X determines strategic sophistication
And X merely related to cognitive ability
Complementary means of investigating effects of Cognitive ability on strategic behavior
5
How to think about the manipulation? Discovered crayon in Homer Simpson’s brain
Was causing cognitive shortcomings
6
Homer without crayon in brainHomer with crayon in brain
How to Manipulate Cognitive Resources? Cognitive Load
Task which occupies cognitive resources Unable to devote to deliberation Observe behavior
7
Cognitive Load in Games Cognitive load and games
Roch et al. (2000) Cappelletti et al. (2008) Duffy and Smith (2012)
Not designed to measure strategic sophistication
Carpenter, Graham, Wolf (2013)
8
Experiment Details 164 Subjects
Rutgers-New Brunswick
Sessions of 16 or 20
Earned average $17.89
z-Tree Fischbacher (2007)
9
Cognitive Load Treatments Before play in each period Subjects given number
to commit to memory Up to 15 seconds
After play in game asked for the number
High Load 9 digit number of 0’s and 1’s First digit always 1
Low Load 3 digit number of 0’s and 1’s First digit always 1
10
Discussion of Cognitive Load Alternate load
High Low
Mandatory rest period 20 seconds between the rounds
Why 0’s and 1’s? Did not want the number to interact with decision
11
Timeline across periods Ten 3x3 games
Pay 3 randomly selected if 10 memorization correct Pay 2 if 9 correct Pay 1 if 8 correct Pay 0 if less than 8
Version of 11-20 Game Version of Beauty Contest
Only paid if memorization correct for both
Randomly matched every period
No feedback
12
Manipulation Checks Load Time
High Load: 10.3 seconds Low Load: 2.4 seconds
p<0.001
Correct Memorization Task Low Load: 98.8% High Load: 97.1%
p=0.004
13
Our version of 11-20 Game Adapted from Arad and
Rubenstein (2012) Subjects are paired Each selects an integer
between 1 and 10 Receive request
10 points = $3.50 Earn a bonus of 10 if
select exactly one lower than opponent
Allows straightforward measure of strategic thinking
Instructions were given before load
Equilibrium 10 w prob. 0.1 9 w prob. 0.2 8 w prob. 0.3 7 w prob. 0.4
14
1-10 Game Response
Bounded above by 10 And below by 1
Run tobit regressions with these bounds
15
1-10 Game Response
16
1-10 Game responses
High Load -0.613(p=0.032)
-0.631(p=0.027)
-0.526 (p=0.087)
Self-reported GPA - - 0.210(p=0.59)
Econ/Game Theory/Female No Yes Yes
Observations 164 164 112
-2 Log Likelihood 603.08 597.84 383.06
High load subjects are more strategic
No evidence that self-reported GPA related to choice
1-10 Game Response
High load more sophisticated! (Not less!!!!!)
High Load subjects expect to face more cognitively able opponent
Low Load subjects expect to face less cognitively able opponent
High Load highlights own limitations
17
Our version of Beauty Contest Every subject selects a half integer
between 0 and 10
Winner guesses closest to 2/3 of average Wins $30
Bounded between 0 and 10
Run tobit regressions with these bounds
18
Beauty Contest Game Response
19
Beauty contest action
High Load 0.667(p=0.085)
0.647(p=0.10)
0.941 (p=0.030)
Self-reported GPA - - -2.079(p<0.001)
Econ/Game Theory/Female No Yes Yes
Observations 164 164 112
-2 Log Likelihood 742.82 739.78 489.32
Relationship between self-reported GPA and strategic sophistication
High load subjects are less strategic
1-10 and Beauty Contest Wrap-up 1-10 game
High load more strategic
Uncomplicated game Play 1 less
than the other guy Self-reported GPA
not related choice
Beauty contest High load
less strategic
Complicated game Guess 2/3
of the average of the session
Self-reported GPA related to choice
20
New Strategic Sophistication Papers Strategic sophistication is the result of
1. Subject’s own ability to make computations
2. Subject’s perception of the strategic sophistication of their opponent
Agranov, Potamites, Schotter, and Tergiman (2012) Alaoui and Penta (2012)
21
1-10 and Beauty Contest Wrap-up Two effects of reduced cognitive resources
1. Reduced ability to make computations
2. Disadvantaged in distribution of cognitive resources of subjects
2 dominates 1 in 1-10 game Relatively uncomplicated
1 dominates 2 in beauty contest game Relatively complicated
22
3x3 Games Always shown as row
player
All payoffs between 1 and 11 points 10 points = $3.50
Unique NE
To specify action And point beliefs of
other’s action
Correct beliefs earn 4 additional points
23
24
Characterize Games Number of own
dominated strategies Of Initial Game
Ranges from 0 to 2
Number of opponent’s dominated strategies Of Initial Game
Ranges from 0 to 2
25
Characterize Strategic Behavior Is the subject selecting the best action
Given that the opponent is selecting each action with prob 0.33?
L1 is not particularly sophisticated
L1 classification behavior 1 if behavior consistent with L1 0 otherwise
26
Repeated measures regressions Regressions
Unstructured covariance matrix within subjects And across observations
Each regression has 1640 observations (164 subjects playing 10 games)
Provide coefficient estimates and p-values
27
3x3 games-L1 Classification
28
LI Classification
High Load 0.136 (p=0.012)
Own DS 0.174(p<0.001)
Opponent’s DS -0.038(p=0.21)
High Load*Own DS -0.108(p=0.007)
High Load*Opponent’s DS -0.0842(0.064)
GPA 0.128(<0.003)
-2 Log Likelihood 1253.3
High load subjects more likely to be classified as L1
High load subjects less sensitive to own dominated strategies
High load subjects more sensitive to opponents dominated strategies
High GPA subjects more likely to be classified as L1
3x3 games-Wrap-up High load subjects
More likely to be L1 Less sensitive to number of own dominated
strategies More sensitive to number of other’s dominated
strategies
29
Conclusion Cognitive load helpful in examining relationship
between cognitive ability and strategic behavior
Pay attention to both Cognitive ability and Perception of cognitive ability of others
30
Future Work Observe timing of the
Action and belief decisions
9 binary digits too easy to remember?
Tell subjects Nash Equilibrium Cognitive resources devoted to behavior of other
Do not solicit beliefs Perhaps prompts subjects to be more strategic
31