choking and excelling under pressure in classification w. todd maddox & arthur b. markman...

Choking and Excelling Choking and Excelling Under Pressure in Under Pressure in

ClassificationClassificationW. Todd Maddox & Arthur B. MarkmanW. Todd Maddox & Arthur B. Markman

University of Texas, AustinUniversity of Texas, Austin

Major Collaborators:

Grant BaldwinBrian GlassLisa GrimmDarrell Worthy

This work is supported by AFOSR grant FA9550-06-1-0204 and NIMH grant R01MH077708

Overview of TalkOverview of Talk Why care about motivation?Why care about motivation?

Pressure (stress) induces a motivational state Pressure (stress) induces a motivational state that affects cognitionthat affects cognition

A framework for thinking about motivation A framework for thinking about motivation and its influence on cognition and learningand its influence on cognition and learning Pressure (stress) effectsPressure (stress) effects

Application to:Application to: Perceptual classification learningPerceptual classification learning Pressure effects on classificationPressure effects on classification

learning and experienced behaviorlearning and experienced behavior

Motivation and CognitionMotivation and Cognition

Why care about motivation?Why care about motivation? No behavior without motivationNo behavior without motivation Motivation affects cognitionMotivation affects cognition Approach positive states and avoid negative Approach positive states and avoid negative

statesstates Motivation rarely controlled in cognitive experimentsMotivation rarely controlled in cognitive experiments

No principled distinctions between motivational No principled distinctions between motivational and cognitive brain regionsand cognitive brain regions

Motivation and Cognition need to be studied Motivation and Cognition need to be studied togethertogether

Motivation and Pressure Motivation and Pressure (Stress)(Stress)

Two current theories of pressure effects on Two current theories of pressure effects on cognitioncognition DistractionDistraction: WM resources reduced: WM resources reduced MonitoringMonitoring: Increased monitoring of explicit : Increased monitoring of explicit

processesprocesses

We suggest that pressure affects cognition We suggest that pressure affects cognition through its effects on the motivational statethrough its effects on the motivational state

Working Hypothesis:Working Hypothesis: Pressure induces an “avoidance” or “prevention” Pressure induces an “avoidance” or “prevention”

motivational statemotivational state

Regulatory Fit Regulatory Fit FrameworkFramework

Reward Structure of Task

Gains LossesPromotion Focus Fit MismatchPrevention Focus Mismatch Fit

Global Motivational Orientation (Higgins: Regulatory Global Motivational Orientation (Higgins: Regulatory Focus Theory)Focus Theory) Approach: Sensitivity to gains in environment (Promotion Approach: Sensitivity to gains in environment (Promotion

focus)focus) Avoidance: Sensitivity to losses in environment (Prevention Avoidance: Sensitivity to losses in environment (Prevention

focus)focus) Pressure associated with Avoidance (prevention) statePressure associated with Avoidance (prevention) state

Local Task Reward StructureLocal Task Reward Structure Gains in the environmentGains in the environment Losses in the environmentLosses in the environment

Regulatory focus interacts with task reward structureRegulatory focus interacts with task reward structure

Consider the bigger Consider the bigger picturepicture

Almost all cognitive research involves a Almost all cognitive research involves a promotion focus and a gains reward promotion focus and a gains reward structurestructure

Recall Pressure associated with Recall Pressure associated with prevention (mismatch)prevention (mismatch)

Reward Structure of Task

Gains LossesPromotion Focus Fit MismatchPrevention Focus Mismatch Fit

Fit

Regulatory FitRegulatory Fit Regulatory focus interacts with task Regulatory focus interacts with task

reward structurereward structure Regulatory fit increases “flexible” Regulatory fit increases “flexible”

cognitive processing.cognitive processing.

Flexibility can be defined within Flexibility can be defined within taskstasks Willingness to test various strategiesWillingness to test various strategies Willingness to explore the environmentWillingness to explore the environment

Hold off question of “why” for nowHold off question of “why” for now

Perceptual Perceptual ClassificatiClassificati

ononMaddox, Baldwin & Markman (2006; Memory Maddox, Baldwin & Markman (2006; Memory

& Cognition)& Cognition)

Perceptual Perceptual Classification TaskClassification Task

Stimuli with small number of Stimuli with small number of underlying dimensionsunderlying dimensions Lines that vary in length, orientation and Lines that vary in length, orientation and

positionposition Experimenter control of category Experimenter control of category

structurestructure Extensive set of tools for modeling Extensive set of tools for modeling

performance of individual participantsperformance of individual participants Can assess the strategies participants Can assess the strategies participants

use in the taskuse in the task

Scatterplot of StimuliScatterplot of Stimuli

o = category A = long, steep lineso = category A = long, steep lines

+ = category B = all others+ = category B = all others

Possible Rule-based Possible Rule-based StrategiesStrategies

83% accuracy100% accuracy

Regulatory FocusRegulatory Focus(Global Task Goal)(Global Task Goal)

Promotion Promotion Focus Focus (Approach)(Approach)

Achieve Global Task Achieve Global Task Performance Criterion Performance Criterion Raffle ticket for $50Raffle ticket for $50

Prevention Prevention Focus Focus (Avoidance)(Avoidance)

Achieve Global Task Achieve Global Task Performance Criterion Performance Criterion Keep $50 raffle ticket given Keep $50 raffle ticket given initiallyinitially

90% correct yields “bonus”90% correct yields “bonus”

Task Reward StructureTask Reward Structure(Local Trial-by-trial Task (Local Trial-by-trial Task

Goal)Goal)

GainsGainsCorrect Response = 3 pointsCorrect Response = 3 points

Incorrect Response = 1 pointIncorrect Response = 1 point

LossesLossesCorrect Response = -1 pointCorrect Response = -1 point

Incorrect Response = -3 pointIncorrect Response = -3 point

Experiment Screen Experiment Screen SampleSample

Gains

Losses

0

18

YesBonus

No

0

18

Correct

21

YesBonus

No

0

21

YesBonus

No

0

22

YesBonus

No

Wrong, that was an A

Experiment Set 1Experiment Set 1Flexibility is AdvantageousFlexibility is Advantageous

Gains Losses

Promotion (try to earn ticket)

Fit: GoodMismatch :

Poor

Prevention (try not to lose

ticket)

Mismatch: Poor

Fit: Good

Conjunctive Rule-Based TaskConjunctive Rule-Based Task Exploration of verbal rule space requiredExploration of verbal rule space required

Performance ResultsPerformance Results

0.7

0.75

0.8

0.85

0.9

0.95

1

Gains Losses

Reward Structure

Pro

port

ion

Cor

rect

Promotion

Prevention

Plots averaged over blocks. Effects generally larger early in learning

ConclusionsConclusions

In a classification task where exploration of the verbal rule space is advantageous, a regulatory fit led to better performance.

Experiment Set 2Experiment Set 2Flexibility is Flexibility is

DisadvantageousDisadvantageous

Gains Losses

Promotion (try to earn ticket)

Fit: PoorMismatch :

Good

Prevention (try not to lose

ticket)

Mismatch: Good

Fit: Poor

Information-Integration TaskInformation-Integration Task

Experimental Method identical to Experiment Set 1

PredictionPrediction If regulatory fit = more flexibility,

then rule-based strategies should persist leading to poorer performance

COVIS assumes that rule-based strategies dominate early..

Rule-based strategies must be abandoned in information-integration tasks

Exceeding the bonus requires abandoning rule-based strategies

Performance Results Performance Results (Losses)(Losses)

0.7

0.75

0.8

0.85

Promotion Prevention

Regulatory Focus

Pro

port

ion

Cor

rect


We observe a 3-way interaction between regulatory focus, task reward structure, and nature of the task.

Flexibility advantageous: Fit is good

Flexibility disadvantageous: Fit is bad

Application Application 2:2:

Choking Choking Under Under

PressurePressureMarkman, Maddox & Worthy (2006; Markman, Maddox & Worthy (2006; Psychological Science)Psychological Science)

Choking Under PressureChoking Under Pressure

Anecdotal phenomenon (e.g. sports, Anecdotal phenomenon (e.g. sports, test-taking, etc.)test-taking, etc.)

People perform worse than normal People perform worse than normal when under pressurewhen under pressure

Might pressure be similar to a Might pressure be similar to a prevention focus?prevention focus?

Categorization TasksCategorization Tasks

Rule-Based Information-Integration

MethodMethod Gains onlyGains only Low pressure – “do your best”Low pressure – “do your best” High pressure:High pressure:

-Paired with a ‘partner’-Paired with a ‘partner’-If both of you reach criterion, both get $6-If both of you reach criterion, both get $6-If one of you fails neither get $6 bonus-If one of you fails neither get $6 bonus-Partner reached criterion-Partner reached criterion

Low-Low-Pressure Pressure Rule-basedRule-based

High-High-Pressure Pressure Rule-BasedRule-Based

Low-Low-PressurePressure

InformationInformation--IntegrationIntegration

High-High-PressurePressure

InformationInformation--IntegrationIntegration

PredictionsPredictions

Low Pressure (Promotion)

High Pressure(Prevention)

Rule-Based Fit: GoodMismatch :

Poor

Information-integration

Fit: PoorMismatch:

Good

Results Results Accuracy

0.54

0.56

0.58

0.6

0.62

0.64

0.66

0.68

0.7

0.72

Information-Integration Rule-Based

Pro

po

rtio

n C

orr

ec

t

High Pressure

Low Pressure

SummarySummary

Pressure does appear to operate like a prevention focus during classification learning (at least with gains).

Pressure (a mismatch with gains) hurts rule-based learning, but helps information-integration learning.

Pressure and Pressure and ExperienceExperience

Method 2500 trials over 4 sessions with NO pressure manipulation 5th session: pressure or no pressure 5th session final block: super pressure

No error in 80 trials yields $100 (no partner)

Categorization TasksCategorization Tasks


Final Control vs. First Final Control vs. First Pressure BlockPressure Block

Final control accuracy high (90% in all conditions Pressure improves II, but not RB (interaction nearly significant)

Difference between the block before and after the pressure manipulation

-0.04

-0.03

-0.02

-0.01

0

0.01

0.02

0.03

0.04


Diff

ere

nce

in A

ccu

racy

Control

Pressure

Final Pressure (or Final Pressure (or Control) vs. Super Control) vs. Super

Pressure BlockPressure Block

Super pressure adversely affected both RB conditions Super pressure adversely affected the II Control condition Super pressure accentuated the II Pressure condition

Difference in accuracy between penultimate and final block

-0.06-0.05-0.04-0.03-0.02-0.01

00.010.020.030.040.05


Category Structure

Diff

ere

nce

in A

ccu

racy

Control

Pressure

SummarySummary

Pressure generally improves II performance even when experienced. Pressure generally hurts RB performance even when experienced.

Related FindingsRelated Findings

Interaction between Regulatory Focus, Interaction between Regulatory Focus, Reward Structure and TaskReward Structure and Task Binary category structures: Same patternBinary category structures: Same pattern Card selection task: Card selection task:

More exploratory behavior for Regulatory FitMore exploratory behavior for Regulatory Fit

May explain some stereotype threat May explain some stereotype threat A stereotype threat induces prevention focusA stereotype threat induces prevention focus Participants perform better under losses Participants perform better under losses

than gainsthan gains

Future ResearchFuture Research

Cognitive mechanisms in this effectCognitive mechanisms in this effect Role of working memory?Role of working memory?

Relationship to individual differencesRelationship to individual differences Personality variablesPersonality variables Cultural difference variablesCultural difference variables

Other tasksOther tasks ForagingForaging Wisconsin Card Sorting TaskWisconsin Card Sorting Task

Thank YouThank You

Super Pressure Block Super Pressure Block First Trial MissedFirst Trial Missed

Final control accuracy high (???%) Equivalent across RB and II conditions Pressure improves II, but not RB

Difference between the first trial missed in the final and penultimate block

-15

-10

-5

0

5

10

15

20


Category Structure

Diff

ere

nce

in fi

rst t

rial m

isse

d

Control

Pressure

Why we should careWhy we should care Motivation affects decision making Motivation affects decision making

Preferences (Brendl, Markman, & Messner; Ferguson & Bargh)Preferences (Brendl, Markman, & Messner; Ferguson & Bargh) Need to smoke increases preference for smoking related items and Need to smoke increases preference for smoking related items and

reduces preference for not smoking related items reduces preference for not smoking related items Goal-adoption (Aarts et al, Fishbach & Shah)Goal-adoption (Aarts et al, Fishbach & Shah)

People adopt goals of people around them. People adopt goals of people around them. Selection of optimal behavior (Bechara et al., Busemeyer & Selection of optimal behavior (Bechara et al., Busemeyer &

Townsend)Townsend)

All cognitive research has an (uncontrolled) motivational All cognitive research has an (uncontrolled) motivational componentcomponent ““motivate” to “try harder”motivate” to “try harder”

““Motivation” brain regions reciprocally connected with Motivation” brain regions reciprocally connected with “cognitive” brain regions“cognitive” brain regions

Regulatory Fit FrameworkRegulatory Fit Framework HypothesisHypothesis: A “fit” between : A “fit” between globalglobal

task goals (regulatory focus) and task goals (regulatory focus) and locallocal task goals (task reward task goals (task reward structure) increases “flexible” structure) increases “flexible” cognitive processing.cognitive processing.

Flexibility can be defined within tasksFlexibility can be defined within tasks Willingness to test various strategiesWillingness to test various strategies Willingness to explore the environmentWillingness to explore the environment

Hold off question of “why” for now?Hold off question of “why” for now?

Regulatory Fit = Flexibility: Regulatory Fit = Flexibility: Why?Why?

Empirical support in several Empirical support in several domainsdomains

Connection to NeuroscienceConnection to Neuroscience Positive affect-frontal dopamine-Positive affect-frontal dopamine-

flexibility hypothesis (Isen, Ashby, flexibility hypothesis (Isen, Ashby, etc)etc)

Regulatory focus-frontal activation Regulatory focus-frontal activation findings (Amodio, Cunningham, etc)findings (Amodio, Cunningham, etc)

LC-NE-exploration/exploitation LC-NE-exploration/exploitation relation (Ashton-Jones, Cohen, Daw)relation (Ashton-Jones, Cohen, Daw)

Rule-Based ModelingRule-Based Modeling

More low pressure (fit) subjects were best fit by More low pressure (fit) subjects were best fit by the rule-based models.the rule-based models.

More high pressure (mismatch) subjects are More high pressure (mismatch) subjects are random.random.

Rule-Based

0%

20%

40%

60%

80%

100%

Low Pressure High Pressure

Per

cent

Bes

t F

it

Random

Info-Int

Rule-Based

Information integration Information integration modelingmodeling

More high pressure subjects best fit by an More high pressure subjects best fit by an information integration model.information integration model.

More low pressure subjects random.More low pressure subjects random.

Information-Integration

0%

20%

40%

60%

80%

100%

Low Pressure High Pressure

Per

cent

Bes

t F

it

Random

Rule-Based

Info-Int

Studying Regulatory Fit Studying Regulatory Fit EffectsEffects

How can we study this systematically?How can we study this systematically? Need task for which we can manipulate Need task for which we can manipulate

the advantageousness of flexibility, while the advantageousness of flexibility, while holding other task characteristics fixedholding other task characteristics fixed

Need a good manipulation of regulatory Need a good manipulation of regulatory focusfocus

Need to be able to manipulate reward Need to be able to manipulate reward structurestructure

Choking Under PressureChoking Under Pressure

Basketball dataBasketball data Free throw during last minute of gameFree throw during last minute of game

Proportion of Free-Throws Made Relative Career Averages

-0.10

-0.08

-0.06

-0.04

-0.02

0.00

0.02

0.04

0.06

-5 -4 -3 -2 -1 0 1 2 3 4 5

Point Margin

Dis

tan

ce fr

om

Ba

selin

e

Model-based AnalysesModel-based Analyses

Decision-bound models (Ashby & Decision-bound models (Ashby & Maddox) fit to each participant Maddox) fit to each participant block-by-blockblock-by-block

Possible Rule-based Possible Rule-based StrategiesStrategies

83% accuracy100% accuracy

Model Fit PredictionsModel Fit Predictions

Regulatory Fit Conditions should be more Regulatory Fit Conditions should be more “flexible” and find the more complex “flexible” and find the more complex conjunctive rule faster than the conjunctive rule faster than the Regulatory Mismatch ConditionsRegulatory Mismatch Conditions

Modeling ResultsModeling Results

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Gains Losses

Reward Structure

Prop

ortio

n C

onju

nctiv

e R

ule

PromotionPrevention

Pro

port

ion

Con

jun

ctiv

e R

ule

Use

Modeling ResultsModeling Results

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Gains Losses

Reward Structure

Prop

ortio

n II

Use

PromotionPrevention

Pro

port

ion

II

Use

Regulatory Fit vs. Loss Regulatory Fit vs. Loss AversionAversion

Fit differs from loss-aversionFit differs from loss-aversion Loss aversion is a main effect. Loss aversion is a main effect.

““Losses loom larger than gains.”Losses loom larger than gains.” Regulatory fit is an interactionRegulatory fit is an interaction

The influence of losses depends on regulatory The influence of losses depends on regulatory focusfocus

Losses lead to Losses lead to flexibleflexible behavior under behavior under preventionprevention

Losses lead to Losses lead to inflexibleinflexible behavior under behavior under promotion promotion

Whether this leads to good or bad overall Whether this leads to good or bad overall performance depends on whether the task calls performance depends on whether the task calls for flexibilityfor flexibility

ApplicatioApplication 2:n 2:

Stereotype Stereotype ThreatThreat

Grimm, Markman, Maddox & Baldwin (under Grimm, Markman, Maddox & Baldwin (under review)review)

Stereotype ThreatStereotype Threat

Stereotype threat is the risk of Stereotype threat is the risk of confirming a negative stereotype confirming a negative stereotype about one’s group (Steele & Aronson, about one’s group (Steele & Aronson, 1995). 1995).

Task-relevant stereotype causes task-Task-relevant stereotype causes task-specific decrements in performancespecific decrements in performance How?How?

Proposed stereotype threat Proposed stereotype threat mechanismsmechanisms

Self-handicapping (Stone et al., 1999; Brown & Josephs, Self-handicapping (Stone et al., 1999; Brown & Josephs, 1999)1999) Blacks better when framed as diagnostic of “natural athletic Blacks better when framed as diagnostic of “natural athletic

ability”ability” Whites better when framed as diagnostic of “sports intelligence”Whites better when framed as diagnostic of “sports intelligence”

Low performance confidence (Cadinu et al., 2003)Low performance confidence (Cadinu et al., 2003) Lower expected level of performance, lower actual performanceLower expected level of performance, lower actual performance

Task-specific concerns (Brown & Josephs, 1999)Task-specific concerns (Brown & Josephs, 1999) Ideomotor priming (Bargh et al., 1996)Ideomotor priming (Bargh et al., 1996)

People primed with elderly stereotype by unscrambling People primed with elderly stereotype by unscrambling sentences relevant to stereotype. Those primed walked more sentences relevant to stereotype. Those primed walked more slowly down hallslowly down hall

Arousal (O’Brien & Crandall, 2003)Arousal (O’Brien & Crandall, 2003) Reduced working memory load (Beilock et al., 2007)Reduced working memory load (Beilock et al., 2007) Regulatory focus (Seibt & Förster, 2004)Regulatory focus (Seibt & Förster, 2004)

Connecting Stereotype Connecting Stereotype Threat and Regulatory FitThreat and Regulatory Fit

Seibt and Förster (2004)Seibt and Förster (2004) Positive stereotype induces promotion; Positive stereotype induces promotion;

negative stereotype induces preventionnegative stereotype induces prevention Promotion induces more elaborative Promotion induces more elaborative

processing while prevention induces more processing while prevention induces more vigilant processingvigilant processing

Others…Others…

FrameworkFramework

GainsGains LossesLosses

Positive Positive stereotype stereotype (“Promotion”)(“Promotion”)

FitFit MismatchMismatch

Negative Negative Stereotype Stereotype (“Prevention”)(“Prevention”)

MismatchMismatch FitFit

Regulatory fit allow for more flexible cognitive processing

Stereotype Threat Stereotype “Threat”?

Flexibility PredictionsFlexibility PredictionsGainsGains LossesLosses


BetterBetter WorseWorse


WorseWorse BetterBetter


Positive Positive stereotype stereotype (“Promotion”(“Promotion”))


Negative Negative Stereotype Stereotype (“Prevention(“Prevention”)”)


Flexibility advantageous (rule-based)

Flexibility disadvantageous (information-integration)

Experiments 1 and 2: Experiments 1 and 2: Flexibility AdvantageousFlexibility Advantageous

Experiment StructureExperiment StructureGainsGains LossesLosses

ExperimeExperiment 1nt 1

Positive Positive stereotypstereotyp

eeWomenWomen FitFit MismatcMismatc

hh

Negative Negative StereotypStereotyp

eeMenMen MismatcMismatc

hhFitFit



eeMenMen FitFit MismatcMismatc

hh


eeWomenWomen MismatcMismatc

hhFitFit

Stereotype Prime Example: Stereotype Prime Example: Women are better in gains Women are better in gains

tasktaskThis is an experiment testing sex differences in This is an experiment testing sex differences in spatial abilities. Previous research has shown that spatial abilities. Previous research has shown that women perform better than men on tests of spatial women perform better than men on tests of spatial ability.ability.

In this experiment, you will earn some points for In this experiment, you will earn some points for correct responses and no points for incorrect correct responses and no points for incorrect responses. In this task, women tend to earn more responses. In this task, women tend to earn more than 86 points per block of trials and men tend to than 86 points per block of trials and men tend to earn fewer than 86 points per block of trials. earn fewer than 86 points per block of trials.

Please try your best in this task. Before continuing, Please try your best in this task. Before continuing, please indicate whether you are male or female. If please indicate whether you are male or female. If you are male, press the "M" key. If you are female, you are male, press the "M" key. If you are female, press the "F" key.press the "F" key.

Task AccuracyTask Accuracy

0.7

0.75

0.8

0.85

0.9

0.95

1

Gains Losses

Reward Structure

Prop

ortio

n C

orre

ct

Women

Men

Experiment 1: Women Positive Stereotype

Experiment 2: Men Positive Stereotype

0.7

0.75

0.8

0.85

0.9

0.95

1

Gains Losses

Reward Structure

Prop

ortio

n C

orre

ctWomen

Men

Prop

ortio

n C

orre

ct

Pro

port

ion

Cor

rect

Model ResultsModel Results

00.10.20.30.40.50.60.70.80.9

1

Gains Losses

Reward Structure

Prop

ortio

n C

J U

se

Women

Men


Experiment 2: Men Positive Stereotype 0

0.10.20.30.40.50.60.70.80.9

1

Gains Losses

Reward Structure

Prop

ortio

n C

J U

seWomen

Men

Prop

ortio

n C

J U

se

Prop

ortio

n C

J U

se


Stereotype threat manipulations led to behavior also observed under a prevention focus. When flexible strategy use is advantageous:

a negative stereotype is disadvantageous in a gains condition. a negative stereotype is ADVANTAGEOUS in a losses condition.







Positive Positive stereotype stereotype (“Promotion”(“Promotion”))


Negative Negative Stereotype Stereotype (“Prevention(“Prevention”)”)


Flexibility advantageous(rule-based)

Flexibility disadvantageous

(information-integration)

Information-integration Information-integration ClassificationClassification

Experiment StructureExperiment StructureGainsGains LossesLosses



eeMenMen FitFit MismatcMismatc

hh


eeWomenWomen MismatcMismatc

hhFitFit



eeWomenWomen FitFit MismatcMismatc

hh


eeMenMen MismatcMismatc

hhFitFit

Task AccuracyTask Accuracy(Preliminary data)(Preliminary data)

0.7

0.75

0.8

0.85

0.9

0.95

1

Gains Losses

Task

Prop

ortio

n C

orre

ct

Women

Men Experiment 3: Men Positive Stereotype


0.7

0.75

0.8

0.85

0.9

0.95

1

Gains Losses

Task

Prop

orti

on C

orre

ct

Women

Men

Positive Stereotype = worse performanceHypothesis supported for gains, but not losses

Preliminary Preliminary ConclusionsConclusions

When flexible strategy use is disadvantageous:

a negative stereotype is ADVANTAGEOUS in gains and losses conditions.

Data collection and modeling is ongoing.


Choice-Choice-Gambling Gambling

TaskTaskWorthy, Maddox & Markman (in press; Worthy, Maddox & Markman (in press; Psychonomic Bulletin and Review)Psychonomic Bulletin and Review)

0

PICK A CARD!YesBonus No

450

174

0

174181

YesBonus No

450

7

0

PICK A CARD!YesBonus No

450

181

0PICK A CARD!

YesBonus No

-450

-311

0

-311-315

YesBonus No

-450

-4

0PICK A CARD!

YesBonus No

-450

-315

Deck CharacteristicsDeck Characteristics Experiment 1: Flexibility (exploration) is good

Disadvantageous deck early becomes advantageous as more samples are taken

Experiment 2: Flexibility (exploration) is bad One deck advantageous


PromotionPromotion BetterBetter WorseWorse

PreventionPrevention WorseWorse BetterBetter

BetterBetterPreventionPrevention

WorseWorsePromotionPromotion

GainsGains

Flexibility advantageous

Flexibility disadvantageous

Flexibility is Advantageous Flexibility is Advantageous Results Results

Average Distance from Criterion

-40

-35

-30

-25

-20

-15

-10

-5

0

GAIN LOSS

Po

ints

Be

low

Cri

teri

on


Flexibility is Flexibility is Disadvantageous Results Disadvantageous Results

Average Distance from Criterion

-100

-90

-80

-70

-60

-50

-40

-30

-20

-10

0

Po

int

Be

low

Cri

teri

on


Model-based AnalysesModel-based Analyses Applied a version of the softmax action selection model to individual subject data (Sutton, Barto, Daw, etc)

Model estimates probability of selecting each deck based on estimated value.

Includes an exploration/exploitation parameter

Modeling ResultsModeling ResultsExploration/Exploitation parameter values

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Gains Losses

Exp

lora

tion-

Exp

loita

tion


Exploration-Exploitation Parameter Values

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Exp

lora

tio

n-E

xplo

itat

ion


Flexibility is Good

Flexibility is Bad

SummarySummary

Regulatory Fit hypothesis applies to choice.

Fit leads to greater exploration.

Extensions to decks that require numerous switches are ongoing.


Signal Signal DetectionDetection

OverviewOverview Two-stimulus identification (line length)

100ms exposure durations, 2 pixel length difference Promotion/Prevention x Gains/Losses Payoffs biased toward “short” mouth

Results: Sensitivity (d’)Results: Sensitivity (d’)

B

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1 2 3 Overall

Block (100-trials per block)

d'

Promotion-Gain

Promotion-Loss

Prevention-Gain

Prevention-Loss

Results: Sensitivity (Results: Sensitivity ())

C

0.6

0.8

1.0

1.2

1.4

1.6

1.8

1 2 3 Overall

Block (100-trials per block)

Bet

a

Promotion-Gain

Promotion-Loss

Prevention-Gain

Prevention-Loss

SummarySummary

Predicted fit to lead to better decision criterion learning.

Prediction unsupported.

Fit increased sensitivity.

Studies with more extensive training are ongoing.

Applications to depressed patients ongoing.

Overall SummaryOverall Summary We outline a framework for understanding the We outline a framework for understanding the

motivation-learning interface.motivation-learning interface. Global: promotion/prevention focusGlobal: promotion/prevention focus Local: gains/lossesLocal: gains/losses Most research involves a promotion focus and Most research involves a promotion focus and

gains.gains. We propose that a fit leads to more flexible We propose that a fit leads to more flexible

(exploratory) processing(exploratory) processing This may or may not be advantageousThis may or may not be advantageous We apply this to several domainsWe apply this to several domains

Classification, choice, signal detectionClassification, choice, signal detection

Open QuestionsOpen Questions Does a fit imply:Does a fit imply:

More efficient exploration of the strategy space, More efficient exploration of the strategy space, oror

Less evidence to abandon a rule?Less evidence to abandon a rule?

Is flexibility always from simple to complex Is flexibility always from simple to complex rules?rules? Bias toward complex rule initially, then gradually Bias toward complex rule initially, then gradually

introduce simple rule that yields bonus.introduce simple rule that yields bonus.

What is the relationship between exploration What is the relationship between exploration in the verbal rule-space and in the card task?in the verbal rule-space and in the card task?

Future DirectionsFuture Directions Understand mental disordersUnderstand mental disorders Many disorders lead to cognitive deficitsMany disorders lead to cognitive deficits

Do these reflect disruptions of information Do these reflect disruptions of information processing?processing?

Perhaps they reflect motivational issuesPerhaps they reflect motivational issues Anxiety disorders may lead to chronic Anxiety disorders may lead to chronic

prevention focusprevention focus Could lead to regulatory mismatchesCould lead to regulatory mismatches

Collaborators/Collaborators/FundingFunding

Darrell WorthyDarrell Worthy

Lisa GrimmLisa Grimm

Brian GlassBrian Glass

Grant BaldwinGrant Baldwin

Supported by NIMH R01 MH77708 and AFOSR grant FA9550-06-1-0204

Summary!!!!!!!!Summary!!!!!!!! Signal Detection Task

Fit increases d’

Choice (Gambling task – ala Iowa Gambling Task)

Fit increases speed to switch to advantageous deck Fit promotes exploration; Mismatch promotes exploitation Worthy, Maddox, & Markman (PB&R, in press)

Performance Pressure Is pressure like a prevention focus (Markman, Worthy & Maddox (Psych Sci, 2006)? How do effects of pressure change with expertise?

Current and Future Current and Future WorkWork

Signal Detection Task

Fit increases d’

Choice (Gambling task – ala Iowa Gambling Task)

Fit increases speed to switch to advantageous deck Fit promotes exploration; Mismatch promotes exploitation Worthy, Maddox, & Markman (PB&R, in press)

Performance Pressure Is pressure like a prevention focus (Markman, Worthy & Maddox (Psych Sci, 2006)? How do effects of pressure change with expertise?

Distraction Hypothesis Distraction Hypothesis

Pressure decreases available Pressure decreases available working memory (WM) resources.working memory (WM) resources.

WM is filled with thoughts about the WM is filled with thoughts about the situation’s importance.situation’s importance.

Negatively influences performance Negatively influences performance on WM demanding tasks on WM demanding tasks

(Beilock & Carr, 2005; Wine, 1971; Beilock et al., 2004, Ashcraft & Kirk, 2001).

Evidence of DistractionEvidence of Distraction

Beilock et al. (2004) - high pressure Beilock et al. (2004) - high pressure reduced subjects’ ability to solve difficult reduced subjects’ ability to solve difficult math problemsmath problems

Beilock & Carr (2005) – high pressure Beilock & Carr (2005) – high pressure caused performance declines for subjects caused performance declines for subjects high in WM capacity, but not for subjects high in WM capacity, but not for subjects low in WM capacity.low in WM capacity.

Ashcraft & Kirk (2001) – individuals with Ashcraft & Kirk (2001) – individuals with high math anxiety did worse on math high math anxiety did worse on math problemsproblems

Explicit Monitoring Explicit Monitoring HypothesisHypothesis

Alternative view of choking Alternative view of choking Pressure causes performance anxiety Pressure causes performance anxiety

which increases the amount of which increases the amount of attention paid to skilled performance attention paid to skilled performance which disrupts performance.which disrupts performance.

Explicit monitoring Explicit monitoring of proceduralized of proceduralized processes disrupts performance.processes disrupts performance.

(Gray, 2004; Beilock, 2001; Masters, 1992; Lewis and Lindner, 1997; Baumeister, 1984)

Evidence of MonitoringEvidence of Monitoring

Masters (1992) – subjects trained with Masters (1992) – subjects trained with explicit golf-putting instructions explicit golf-putting instructions performed worse than those w/o explicit performed worse than those w/o explicit instructionsinstructions

Gray (2004) – Expert batters performed Gray (2004) – Expert batters performed worse when focusing on their technique.worse when focusing on their technique.

Lewis and Linder (1997) – found Lewis and Linder (1997) – found evidence supporting monitoring and evidence supporting monitoring and disconfirming distraction in golf-puttingdisconfirming distraction in golf-putting

How Does Pressure Affect How Does Pressure Affect Category Learning?Category Learning?

Multiple Systems Proposal of Category Multiple Systems Proposal of Category Learning (e.g. Maddox and Ashby, Learning (e.g. Maddox and Ashby, 2004)2004)

Different neural circuits mediate Different neural circuits mediate learning of different category structureslearning of different category structuresExplicit, hypothesis testing system mediates learning Explicit, hypothesis testing system mediates learning of “rule-based” (RB) category structures.of “rule-based” (RB) category structures.Implicit, procedural learning system mediate learning Implicit, procedural learning system mediate learning of of “information-integration” (II) category “information-integration” (II) category structures.structures.

Rule BasedRule Based Verbalizable rule Verbalizable rule

can separate the can separate the two categoriestwo categories

Requires working Requires working memory resources memory resources to learn the to learn the category structurecategory structure

Dual WM task Dual WM task interferes with interferes with learning learning (Zeithamova & (Zeithamova & Maddox, 2006)Maddox, 2006)

Verbalizable

Information IntegrationInformation Integration

Cannot be solved by Cannot be solved by a verbalizable rule.a verbalizable rule.

Rule use may Rule use may interfere with interfere with learninglearning

Dual WM task does Dual WM task does not interfere with not interfere with learning learning (Zeithamova & (Zeithamova & Maddox, 2006)Maddox, 2006)

Not verbalizable

Predictions from Theories of Predictions from Theories of ChokingChoking

Distraction Distraction and and Monitoring Monitoring theories make opposite theories make opposite predictions.predictions.

DistractionDistraction – Pressure should harm – Pressure should harm performance on rule-based tasks, but performance on rule-based tasks, but should have either no effect or a positive should have either no effect or a positive effect on information-integration taskseffect on information-integration tasks

Monitoring Monitoring – Pressure should harm – Pressure should harm performance on information-integration performance on information-integration tasks, but should have either no effect or a tasks, but should have either no effect or a positive effect on rule-based taskspositive effect on rule-based tasks

Monitoring

Distraction

Distraction Theory's Prediction

0.45

0.5

0.55

0.6

0.65

0.7


Pro

po

rtio

n C

orr

ec

t

High Pressure

Low Pressure

Monitoring Theory's Prediction

0.45

0.5

0.55

0.6

0.65

0.7


Pro

po

rtio

n C

orr

ec

t

High Pressure

Low Pressure

?

?

HypothesesHypotheses

Distraction – Pressure should harm performance on rule-based tasks due to depletion of working memory resources, but should have either no effect or a positive effect on information-integration tasks

Monitoring – Pressure should harm performance on information-integration tasks due to a focus on explicit strategies, but should have either no effect or a positive effect on rule-based tasks

Experiment 1 DetailsExperiment 1 Details

Figure 3. (a) Proportion of participants who exceeded the performance criterion in Experiment 1a (gains). (b) Proportion correct (averaged across observers) from Experiment 1a along with standard error bars. (c) Proportion of participants whose data was best fit by a conjunctive model in Experiment 1a. (d) Proportion of participants who exceeded the performance criterion in Experiment 1b (losses). (e) Proportion correct (averaged across observers) from Experiment 1b along with standard error bars. (f) Proportion of participants whose data was best fir by a conjunctive model in Experiment 1b.

Flexibility is BadFlexibility is Bad

(b)

0.400.500.600.700.800.901.00

Attainable Unattainable

Goal State

Pro

por

tion

Cor

rect

Promotion

Prevention

25

30

35

40

45

50

Attainable UnattainableInte

rnal

Noi

se S

tan

dar

d D

evia

tion

Promotion

Prevention

II taskII task

Figure 6. (a) Proportion correct (averaged across observers) from Experiment 3 along with standard error bars. (b) Proportion of participants best fit by an II model. (c) Proportion of participants best fit by a CJ model. (d) Proportion of participants best fit by a UD model. (e) Proportion of participants best fit by a Random responder model.

SDT TaskSDT Task

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1.40

1.60


Situational Focus

d' Gains

Losses

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1.40

1.60


Situational Focusb

eta Gains

Losses

choking and excelling under pressure in classification w. todd maddox & arthur b. markman...

Documents

cognitive research

environmentregulatory

category b

afosr grant fa9550

cognitionwhy care

austinmajor collaborators

wm resources reducedmonitoring

experienced behaviormotivation