Working memory and intelligence, looking at its relationship through Brunswiks lens

1
Working memory and intelligence, looking at its
relationship through Brunswiks lens
Werner W. WittmannUniversity of Mannheim,
Germany Symposium Working memory and
intelligence Controversy or consensus?
organized by Phillip L. Ackerman, Georgia
Institute of Technology Presented at APS - 15th
Annual ConventionAtlanta, GA - May 29 June
1, 2003
2
Clinical prediction paradigm schematized by
Brunswiks lens model(After Hammond, Hursch, and
Todd, 1964)
3
Thank you Egon and congratulations to your 100th
anniversary
Egon Brunswik 1903-1955 (Photo courtesy of
Department of Psychology,University of
California, Berkeley)
4
THE HIERARCHICAL BRUNSWIK LENS MODEL
PREDICTOR AREA
CRITERION AREA
5
The Berlin Model of Intelligence (BIS)
CONTENT
OPERATION
Each parcel is build as an aggregate of three
z-scored tasks, e.g.Sparcel1 (ZBD ZTG ZXG)
6
Intelligence and many unanswered questionsBerlin
Model of Intelligence (BIS-4 test)
7
The set of working memory tasks in the WMC95-study
Working Memory Capacity
verbal
Reading spanVerbal span
Contents
Spatial-figurative
Spatial WMPattern Transform.MU Spatial STM
Alpha SpanVerbal Coord
numerical
TrackingSpatial CoordSpatial IntegrationMU
spatial STM Spatial STM
Comp. SpanBackward digit spanMU numericalMath
span
Switching verbalCategory gen.
Storage Processing
Switching figurativeRandom gen.
MU numerical
Coordination
GaußSwitching numericalStar counting
Functions
Supervision
8
BERLIN MODEL OF INTELLIGENCE
WORKING MEMORY MODEL (WMC_95)
LEVEL OFGENERALINTELLIGENCE
LEVEL OF GROUP FACTORSOF INTELLIGENCE
OPERATIVEAND CONTENTMODE
CELLLEVEL
SINGLE TASKLEVEL(BIS-4)
SINGLE WORKINGMEMORY TASK
LEVEL OF ORTHOGONALIZED WORKING-MEMORY
GROUP-FACTOR
LEVEL OF GENERAL WORKING MEMORY FACTOR
MF
F
Short TermMemory (F)
MN
Content Factorgc ?
.860
N
WMC-SPAT
MV
Memoryupdating (F)
.656
.670
SF
.890
V
Spatial (F)coordination
SN
WMC-g
g
ReadingSpan (V)
SV
.846
M
.600
WMC-NV
CF
Computation Span (N)
.829
CN
S
Operative Factorgf ?
.588
Memoryupdating (N)
CV
.438
C
RF
Switching (N)
.592
WMC-Switching
R
RN
Switching (V)
.845
.814
RV
Switching (F)
9
Relating the three WMC-group factors to the
operative factors (WMC95_study)
WORKING MEMORY GROUP FACTORS
BIS-OPERATIVE GROUP FACTORS
E 4
.74
.33
WMC-NV
R20.45
.33
.48
-.39
.24
E 5
.97
R20.06
-.48
WMC-SPEED
.25
-.30
E 6
.93
.26
R20.13
.20
.49
WMC-SPAT
-.17
E 7
.83
R20.31
10
Relating the BIS-operative group factors to the
three WMC-group factors (WMC95_study)
WORKING MEMORY GROUP FACTORS
BIS-OPERATIVE GROUP FACTORS
.37
WMC-NV
E 1
.80
R20.35
.28
.48
.36
.47
WMC-SPEED
E 2
.23
.81
R20.35
.17
-.44
.28
.19
.49
WMC-SPAT
E 3
.81
.16
R20.34
11
Facet Taxonomy for Working Memory Tasks WMC_97
study
Notes The first column represents the
hypothesized components involved in the tasks.
CRT choice reaction time task.
12
Relating the three WMC-group factors to the
operative factors (WMC97_study)
WORKING MEMORY GROUP FACTORS
BIS-OPERATIVE GROUP FACTORS
E 1
.92
.40
WMC 97-SP
R20.16
.30
-.27
.24
-.18
E 3
.77
R20.41
.52
WMC 97-CO
-.26
-.22
.18
E 4
.92
R20.15
-.21
-.23
.21
WMC 97-PROC
E 2
.98
R20.03
13
Schmid-Leiman models for working memory and
intelligence (WMC_97)
.61
CO1V
ROP1
E 1
.36
.51
Co_res
REAS_res
CO1N
ROP2
.57
E 2
.58
.39
.24
.38
COF2
ROP3
E 3
D 4R21.00
.60
MOP1
SPV
E 5
.29
.30
1.00
g
wmcg
MEM_res
SPN
MOP2
E 6
MOP3
SPF2
E 7
.40
.54
PROCV
SOP1
E 8
.59
.59
Pro_res
PROCN
Speed_res
SOP2
E 9
.45
.66
.68
.42
.17
PROCF2
SOP3
E 11
Chi sq.184,71 P0.00 CFI0.95
RMSEA0.06
14
Disentangling (Cor)relationships in the
Brunswik-Symmetry framework WMC95-study
MF
Factors NON-R used as THEORY-DERIVED SUPPRESSORS
MN
MV
WMC-NV
SF
M
g
SN
S
SV
WMC-SPAT
WMC- g
Operative Factor gf?
C
CF
CN
R
CV
WMC-SPEED
RF
RN
RV
15
Experimenting with theory derived suppressors in
the Brunswik-symmetry framework
Reasoning and WMC-g R .648 R2 .419
adj. R2 .415
Reasoning and WMC-gwith non-Reas () factors as
suppressors R .712 R2 .507 adj.
R2 .492 Increment dR² .088
Reasoning and WMC group factors R .669 R2
.447 adj. R2 .435
Reasoning and WMC group factors with non-Reas ()
factors as suppressors R .724 R2 .524
adj. R2 .502 Increment dR2 .077
N 135
PREDICTOR WMC-g
STD COEF .648
STD COEF .803-.234-.173-.175
PREDICTOR WMC-NVWMC-SPATWMC-SPEED
STD COEF .495.422.155
STD COEF .581 .494 .269-.207-.184-.171
All beta-weights plt.01
The results demonstrate higher symmetry at a
lower level of generality. This tool is very
helpful for differential-correlational research
in explaining and understanding relationships
between constructs.
16
Testing working memory and intelligence at
real-life criteria
Model of performances in complex computer based
business gamesopi_g General problem solving
capacity
General(g) and not(g) factors as intelligence as
process factors from the WMC95_study
Chi sq.18.68 P0.95 CFI1.00
RMSEA0.00
17
Testing Ackermans PPIK-theoryWMC95_study
Chi sq.28.51 P0.87 CFI1.00
RMSEA0.00
18
Reasoning is a little bit more than working
memory capacity?!Wmc97_study
Chi sq.52,76 P0.12 CFI0.98
RMSEA0.04
19
Summary and conclusions

• The relationship between working memory and
  intelligence is close but not identical.
• It depends pretty much on what is defined as a
  working memory task and as an intelligence task.
• Experimental cognitive research too often uses a
  single task only, thus the problem of level of
  generality is rarely visible and dealt with, odds
  for mismatch and asymmetry are high and the
  generalization of results is heavily endangered.
• Differential cognitive researchers favorite
  g-factor at one level of generality easily can be
  a more circumscribed group factor at a more
  general level.
• Principles of Brunswik-symmetry can help in
  improving prediction and explanation while
  searching for better understanding of construct
  validity. Symmetry is a key concept in all
  successful sciences.Thank you Egon Brunswik!