Measuring working memory capacity and cognitive controlcntrics.ucdavis.edu/meeting2/7Engle.pdf · Measuring working memory capacity and cognitive control ... analogy readcomp remoassoc

10/30/07 CNTRICS St. Louis 2007 1

Measuring working memorycapacity and cognitive control

Randall W. Engle

Georgia Institute of Technology

All papers available at http://psychology.gatech.edu/renglelab/


Outline

My question

My approach

My thoughts on WM as a system and specifically on WMC

Complex span tasks that ‘work’.Construct and predictive validity of WMC tasks

Evidence that WMC reflects individual differences in cognitive control

Take home message


My initial question:

Why do complex span tasks predictperformance on higher-level tasks?

What is the nature of the mediating construct?


Combination of psychometric and experimentalmethods. Cronbach (1957) and Underwood(1975).

Benefits and Limitations

My approach is inductive rather than deductive.

My approach


My thoughts on working memory capacity:

Focus on limits in the NUMBER of items (7±2 or 4±1) is not helpful.Problem is in the interface between attention and retrieval frominactive memory and in the prevention of attention capture byrepresentations not relevant to the current task.Much of what we need to know to function, even in the modernworld, can be derived from largely automatic retrieval from LTMwithout need of WM.However, proactive interference can make automatic retrieval aproblem – when the wrong action is retrieved. So we need to havesome way to keep important information either in the primary memoryor easily accessible from secondary memory. – An American driving orcrossing the street in England.


Other thoughts on WMC and Cognitive Control

Load from any source that captures attention can diminish thatcapability (e.g., intrusive thoughts and emotions). Load can turnhigh spans into low spans.

The construct is important to regulation of thought, emotion, andbehavior.

Question: Are cognitive declines associated withpsychopathology a result or a cause?


Simple Span Reading Span Operation Span______ (WMC) (WMC)

B The tiger leapt to the ridge. B Is (3 x 1) – 1 = 3 ? B

N I’ll never forget my days of combat. N Is (10 / 2) + 1 = 6 ? N

K Andy was arrested for speeding. K Is (8 / 4) – 1 = 1 ? K

J The mirror cast a strange reflection. J Is (3 x 3) + 1 = 12 ? J

S Broccoli is a good source of nutrients. S Is (4 x 3) + 2 = 14 ? S

Verbal STM and WMC Tasks


Simple span and complex spans reflect differentconstructs

Engle et al (1999, JEP:G)

OSPAN

RSPAN

CSPAN

BSPAN

FSPAND

FSPANS

.64

.78

.79

.74

.61

.70

WM

STM

gF

RAVENS

CATTELL

.77*

.63*

.61*

.67*

.79*

.71*

.81*

.85*

.29*

.91*

.74*

.41

.67

.49*

.12

COMMON

error

error


Crystallized abilities

Fluid abilities


Spatial WMC Tasks

???

???

???

Rotation-Arrow Span (set sizes 2 - 5)

Symmetry-Matrix Span (set sizes 2 - 5)

Navigation-Ball Span (set sizes 2 - 5)

G F


WMC Construct is domain-general Engle et al (1999) and Kane et al (2004)

ReadSpan

NaviSpan

SymmSpan

CounSpan

OperSpan

RotaSpan

WM

.71

.73

.69

.82

.78

.84

ReadSpan

NaviSpan

SymmSpan

CounSpan

OperSpan

RotaSpan

WM-V

.75

.72

.76

.82

.79

.85

WM-S

Model A1 Model A2

.93

Χ2/df=1.72, CFI=1.0, NFI=.99,NNFI=.99, RMSEA=.02 Χ2/df=1.72, CFI=1.0, NFI=.99,NNFI=.99, RMSEA=.02


Kane et al (2004)wordspan

lettspan

digtspan

operspan

readspan

counspan

symspan

rotaspan

ballspan

navispan

arrospan

matxspan

Storage-V

ExecAttn

Storage-S

inference

analogy

readcomp

remoassoc

syllogism

ravens

beta3

spacerela

rotablock

wasi

surfdevel

formbord

paperfold

Reason-V

Gf

Reason-S

.52*.29*

.25*


WMC

operspan

readspan

counspan

Storage

word1a

word1b

word2aGf

ravens

cattell

PS

letter

pattern

digitsym

word2b

.60*

.07

.18

Conway et al (2002)


WMcapacity

Higher-order tasks

Lower-order tasks

Reasoning

Process dissociation (C not A)

Vulnerability to PI

Dichotic listeningStroop

Object-based visual attention

Mind wandering

Antisaccade

Complex learning

Writing

Note taking

Vocabulary learning

Following directionsReading/listening comprehension

Don’t think of ‘white bears’ – intrusive thought

0

20

40

60

80

100

High Span Low Span

Perc

en

t R

ep

ort

ing

Nam

e

Conway, Cowan & Bunting

REDBLUE

GREENGREEN

REDGREENBLUERED

GREENRED

BLUEBLUE


WMcapacity

Reasoning

Process dissociation (C not A)

Vulnerability to PI


Object-based visual attention

Mind wandering

Antisaccade

Complex learning

Writing

Note taking

Vocabulary learning


Don’t think of ‘white bears’WMC as Trait Variable

Brain structures

Neurotransmitters

Genes


WMcapacity

WMC as State Variable

Drugs(Peter Finn’s work)

Life Event Stress(Kitty Klein’s work)

Senile DementiaAltzheimer’s Type(Ginny Rosen’s work)

Sleep deprivation(Pilots and medical residents)

Stereotype threat(Toni Schmader’s and

Jennifer Richeson’s work)

Choking under pressure(Sian Beillock’s work with math problems)



Antisaccade

Process dissociation (C not A)Vulnerability to PI

Object-basedvisual attention


WMC

ReasoningComplex learning

WritingNote takingVocabulary learning

Don’t think of ‘white bears’

Individual Differences

Sleep deprivation

Drugs

Life Event Stress

Stereotype threat

SDAT

WMC as mediating construct


Evidence that differences in WMC correspond to attentioncontrol:

Antisaccade TaskEriksen TaskAttention network task

The antisaccade task is a model for attention control and the regulation ofhighly prepotent behavior – behavioral inhibition.

Unsworth, Schrock & Engle (2004)


Unsworth et al. (2004)

Trial Example (half speed) - watch the cross hairs!!


Unsworth et al. (2004))

0

2

4

6

8

10

12

14

Prosaccade Antisaccade

Trial Type

Pe

rc

en

t e

rro

r

Low span

High span


Heitz and Engle (2007 JEP:G)

Is WMC as measured by complex span tasks related to the speedof attention control?

Response deadline procedure (700-200 msec)


Eriksen flanker task

Press left key if target is S and right key if target is H

.SSSSSHHHHHHHSHHSSHSS

Response deadline 200-700 msec


HHSHH

HHSHH

HHSHH

Focus of attention as a function of time

time


Heitz & Engle

Incompatible Bins - Low Discriminability

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 100 200 300 400 500 600 700 800

RT

Pro

po

rtio

n C

orr

ec

t

LS

HS

Look at accuracy as function of Rt bins

No difference between Hi and Lo WMC on compatible trials


The Rate of Attention Control Model

High Span

S S H S S

S S H S S

S S H S S

S S H S S

S S H S S

S S H S S

Low Span

TIM

E

S S H S S

S S H S S

S S H S S

S S H S S

S S H S S

S S H S S


HS/LS may not differ in the ability to constrainattentional field per se?

They do differ in the speed of that ability (rate ofattentional constraint)


Value of combination of Experimental and Psychometricapproaches – Cronbach (1957)

1. Two WM tasks Ospan and Nback. Presumably they both measure thesame thing. Kane and Conway showed that both accounted forvariance in Raven but little of it was common.

2. One task, Stroop, clearly two versions measure different things. Onlyget WMC differences on incongruent trials in presence of congruenttrials – none with only incongruent trials.

3. Even subtle variation in a task will lead to big changes in what a taskmeasures. Unsworth & Engle (2006) Simple span and list length.

4. Distinction between Stability and Flexibility or Maintenance andUpdating.


Complex span and gF

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

2 3 4 5

List Length

Co

rre

lati

on

wit

h g

F

Series1


Simple span and gF

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

2 3 4 5 6 7

List-Length

Co

rre

lati

on

wit

h g

F


Michael Posner’s Attention Networks - three ‘independent’ attention networks

Alerting – general readinessOrienting - to specific region of space or timeExecutive Control/Conflict

– Resolving conflict among responses– Associated with conflict tasks such as

Stroop and flanker


Redick and Engle (2006)

Predict IDs in WMC would correspond toExecutive network but not sure about alertingor orienting– Different WMC Ss differentially control

attention on incompatible vs. compatibletrials


Redick and Engle (2006)

31.49

63.12

121.57

40.63

51.30

88.36

0

10

20

30

40

50

60

70

80

90

100

110

120

130

140

Alerting Network Orienting Network Executive Network

Attentional Networks

Dif

fere

nc

e s

co

re (

ms

)

Low Span High Span

No Cue― Double Cue

Alerting

Center Cue― Spatial Cue

Orienting

Incompatible― Compatible

Executive


What we have called ‘working memory capacity’ reflectsindividual differences in ability to control the contents of thefocus of attention and plays a crucial role in the regulation ofcognition, emotion, and behavior.

I believe that construct reflects some aspect of functioning ofnormal frontal executive areas of the brain (PFC, anteriorcingulate).

There is growing evidence for a role of genetics in this system.

WMC construct is important for:(1) rapid access to task-relevant information particularly

in the presence of distraction and/or interference,(2) suppressing potentially interfering information AND

maintaining access to information, and(3) in the flexibility of attention allocation.

Take home message


All papers available in downloadable pdf athttp://psychology.gatech.edu/renglelab/

Tasks available from same web site but you mustemail webmaster for password


0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

A-X A-nonX nonA-X nonA-nonX

Trial Type

Err

or

Ra

te

Low Span Short Low Span Long High Span Short High Span Long


Sleep deprivation


Effects of sleeplessness (30-36 hours) on flight performance

Gyroflight Sustained Operations Simulator (GSOS)

• Features 4-axis simulator motion and 3-channel visualsystem

• Models T-6 aircraft• Unique capability of presenting motion and visual

conflicts during an almost entirely automatedsimulator flight profile


10 pilots studied (run in tandem)– 34.2 yrs (23-46)– 2806 flt hours (207-5800)

Pilot schedule– Training on evening of Day 1 (3 training flights)– Arrive at 0730/0800 on Day 2– Begin first session at 1200/1300– Continuous wakefulness until 1630/1730 on Day 3 (10

sessions)

Basic Methodology

1230 1530 1830 2130 0030 0330 0630 0930 1230 1530

Time of Session


Difficulty of finding a task that predicts sleep-relatederror

Perceptual Vigilance task- Previous gold standard of fatigue

0000

Correlation (flight performance and PVT)=.44

Correlation (flight performance and OSPAN)=-.65

Results


Schmader & Johns (2003)


Beilock and Carr (2005)

Documents

Measuring working memory capacity and cognitive controlcntrics.ucdavis.edu/meeting2/7Engle.pdf · Measuring working memory capacity and cognitive control ... analogy readcomp remoassoc