Belief bias in conditional reasoning - an fMRI study.

Belief bias in conditional reasoning - an fMRI study.
Matt Roser, Jonathan Evans, Nick McNair, Giorgio Fuggetta, Dries Trippas
RES-062-23-3285
Belief bias in conditional reasoning - an fMRI study.
Jonathan Evans
Nick McNair
Giorgio Fuggetta
Dries Trippas
Marie-Stephanie Cahart
Lucas Michaelides
Belief bias in conditional inference
• Many cognitive biases influence reasoning and decision making
• Performance on reasoning tasks often diverges from normative correctness
but at least some people can reason analytically on these tasks
• We studied Belief Bias in conditional reasoning. This is manifest as a tendency
to endorse more inferences derived from believable conditional statements,
regardless of logical validity
• In the dual-process framework – type 1(intuitive) processes cause the belief
bias which may or may not be overridden by type 2(reflective) processes
• Individuals of higher cognitive ability (and WMC) are more likely to inhibit the
effects of believability in conditional reasoning, under the right conditions
Method and research questions
• We studied conditional reasoning while scanning with fMRI
- In addition, we collected IQ, working-memory and measures of
rational thinking disposition from all of our participants
• Research questions
-
what neural systems are involved in conditional reasoning?
-
how is believability processed in the brain?
-
what distinguishes those more or less susceptible to belief bias in terms of
neural activity?
-
in particular, do those showing less belief bias, show more evidence of
inhibition and/or more evidence of engagement of analytic reasoning
Brain areas of interest
• Right Inferior Frontal Gyrus (IFG) – inhibition
Goel, & Dolan (2003). Cognition, 87(1), 11-22.
• Dorsolateral prefrontal cortex (DLPFC) – working memory
• Ventromedial cortex – belief
• VLPFC and connected posterior (temporal) areas
Barbey, Koenigs & Grafman (2013). Cortex,
49(5), 1195-1205.
O’Reilly, R. C. (2010). Trends in neurosciences,
33(8), 355-361.
Harris, Kaplan, Curiel, Bookheimer, Iacoboni, &
Cohen (2009). PLoS One, 4(10), e7272.
Stimuli and Task
• Three-term conditional statements comprised the Major Premise (MajP),
Minor Premise (minP), and Conclusion (C)
• Conditionals described causal relationships about the world
• Half of the MajP statements were ‘believable’, half ‘unbelievable’
Believable MP
If car ownership increases then traffic
congestion will get worse
Car ownership increases
Therefore, Traffic congestion gets worse
--Believable DA
If jungle deforestation continues then gorillas
will become extinct
Jungle deforestation does not continue
Therefore, gorillas do not become extinct
Unbelievable MT
If fast food is taxed then childhood obesity will
increase
Childhood obesity does not increase
Therefore, fast food is not taxed
--Unbelievable AC
If the lottery prize-money increases then fewer
people will buy tickets
Fewer people buy tickets
Therefore, lottery prize-money increases
Stimuli and Task
•
N = 51
•
72 x experimental trials in 3 x blocks of 24
– Half Un/believable
– Half In/valid
– Each MajP was shown twice: 1 x Valid and 1 x Invalid
•
36 x baseline-task trials in 2 x blocks of 18
Exp
BL
Exp
BL
Exp
A trial
+
Major
Premise
Major
Premise
Minor
Premise
Confidence
+
Response
3-5
sec
Conclusion
1
sec
4-6
sec
Response
• Split presentation allows separation of the sentence containing the belief
content (the major premise) from the content that allows reasoning to begin
(the minor premise and conclusion)
Instructions
Reasoning Condition:
• Participants shown an example
• Assume the premises are true
• Determine whether the conclusion must necessarily follow by logic
• Two-button choice response – 15 seconds
Baseline Condition:
• Participants shown an example - three unrelated premises
• Check for repetitions of verbs – one versus none or two
• Two-button choice response – 15 seconds
Baseline statements
“YES”
If jungle deforestation continues then gorillas will become extinct
Foreign investment is encouraged
Therefore, Prince Charles will become the king
“NO”
If fertility treatment improves then the world population will rise
Sea levels rise
Therefore, business practices will improve
Acceptance Rates: Overall
• A tendency to treat inferences from believable conditionals as logically
valid was observed
Believable
Unbelievable
MP
97.6
90.1
MT
43.7
34.3
AC
43.1
36.6
DA
75.6
61.5
Overall
65.0
55.6
Belief: F(1,49)= 25.80, p< .001
Form: F(3,147)= 80.24, p< .001
MP > DA > AC = MT
100
90
80
70
60
50
Believable
40
Unbelievable
30
20
10
0
MP
MT
AC
DA
Overall
Acceptance Rates: Low / High IQ
Belief*IQ: F(1,49)= 5.35, p= .025
Low IQ: Believable > Unbelievable
High IQ: Believable = Unbelievable
Form*IQ: F(3,147)= 6.53, p= .001
Low IQ: MP > DA = AC = MT (DA > MT)
High IQ: MP > DA > AC = MT
100.0
100.0
90.0
90.0
80.0
80.0
70.0
70.0
60.0
50.0
Believable
40.0
Unbelievable
30.0
60.0
50.0
Believable
40.0
Unbelievable
30.0
20.0
20.0
10.0
10.0
0.0
MP
MT
AC
Low IQ
DA
Overall
0.0
MP
MT
AC
High IQ
DA
Overall
IQ and Belief Bias
50
40
Belief Bias
30
20
Pearson’s r: -.31 (p< .05)
10
0
80
90
100
110
-10
-20
WASI IQ Score
120
130
fMRI Data Model
BOLD modelled for major premise (P1)
and for entire argument (PALL)
+
Major
Premise
Major
Premise
Minor
Premise
Confidence
+
Response
3-5
sec
Conclusion
1
sec
4-6
sec
Response
Processing and Analyses
•
•
•
•
•
•
•
High-pass filter – cutoff 128s
Smooth with a 6mm FWHM kernel
Normalisation to MNI template
Six motion regressors were included (3x translation; 3x rotation)
Temporal derivatives not included
Four subjects excluding from analysis due to functional/movement artifacts
Three subjects excluded on behavioural grounds (accepted or rejected all)
• Reasoning Task > Control for Major premise (P1) and Entire Argument (PAll)
• Contrasted Believable/Unbelievable for Major premise and Entire Argument
• Correlate Belief-Bias index with extracted signal strength for each ROI
•
Correction for Multiple Comparisons
– Voxel-level: p < .001 uncorrected
– Cluster-level: p < .05 FDR corrected
P1 > Control Task
Left Middle Temporal gyrus – Temporal pole
Left Inferior Parietal cortex - Angular gyrus
Left Inferior Frontal cortex (BA 44)
R Hemisphere homologues show lesser activity
Left SMA, Dorsomedial PFC
Bilateral ventromedial frontal
Right Inferior frontal (BA 45)
P1: Unbelievable > Believable
Left and Right
Inferior Frontal Gyrus
Right more ventral
BA 45/47
Left TPJ
P1: Believable > Unbelievable
Ventral and Dorsal
Anterior cingulate
PAll > Control Task
Left Dorsolateral PFC
BA 44
BA 8 and BA 9
PAll: Unbelievable > Believable
Left and right middle
frontal gyrus BA 44/45
Left and right parietal
lobe
Belief Bias: PAll Unbelievable > PAll Believable
Index of belief effect: Believable – Unbelievable endorsement rates
Left MFG
0.5
0.4
0.3
Pearson’s r: .613 (p< .001)*
0.2
0.1
0
-20
-0.1 0
20
40
60
-0.2
-0.3
Left PPC
0.6
0.5
0.4
0.3
Pearson’s r: .512 (p< .001)*
0.2
0.1
0
-20
-0.1 0
-0.2
20
40
60
Belief Bias: PAll Unbelievable > PAll Believable
Index of belief effect: Believable – Unbelievable endorsement rates
Right MFG
0.5
0.4
0.3
Pearson’s r: .642 (p< .001)*
0.2
0.1
0
-20
-0.1
0
20
40
60
-0.2
Right PPC
0.4
0.3
0.2
Pearson’s r: .591 (p< .001)*
0.1
0
-20
0
-0.1
-0.2
20
40
60
• Lateral inferior-frontal areas were differentially active according to the
believability of the major premise
• Involvement of representational and inhibitory mechanisms at an early
stage of the trial invoked by the presentation of belief-laden material when
participants are instructed to reason deductively.
• Dorsolateral prefrontal and parietal regions were differentially activated by
believability when all premises available for inference
• Individual differences in the degree of belief bias correlated strongly with
functional activation in DLPFC and PPC at the later stage of the trial.
• Individual differences in the degree of belief bias are reflected in later
processes of premise integration which draw heavily on working memory.
• Areas show greater activity with unbelievable materials – demands on
them are greater when reasoning with unbelievable material.
• People who are more susceptible to belief bias show a greater degree of
differential demand on these regions when believability is manipulated.
• These results are consistent with dual-process accounts of reasoning.
They imply an initial processing of belief that is actively inhibited
(according to instructions) but which can determine responses if WMdemanding processes are insufficient
Believability modulated activity in IFG/Temporal network and
DLPFC/Parietal network under different stimulus conditions (P1,PAll)
Locus of BB effect in latter stage of premise integration and manipulation
(PAll), which draws heavily on WM and activates DLPFC/PPC, consistent
with negative correlation between IQ/WM and BB.
Inference
follows
No
Reinstate
Yes
Yes
Believable?
Check
reasoning
Draw
Withhold
Yes
No
Draw
Withhold
No
Withhold
Reinstatement of inference in opposition to belief
Greatest in high IQ, leading to less BB
Low IQ experience relatively greater demand on WM resources when reasoning
with unbelievable materials and are less able to reinstate the valid inference
Within-trial rTMS in Conditional Reasoning
Within-trial rTMS in Conditional Reasoning
• Investigate stages of Major-Premise (MajP) processing (believability apparent) and
Minor-Premise (minP) (Between subjects)
• Investigate functional regions such as DLPFC (working memory) and IFG (inhibition),
informed by group fMRI de-normalised to individual
Cross these two factors in a 2x2 design:
MajP – IFG
MajP – DLPFC
minP – IFG
minP – DLPFC
• MajP - Decontextualisation of materials and the suppression of belief may occur
following the MajP.
• minP - Effortful combination of premises and the derivation of a conclusion
• An effect of rTMS over IFG for MajP, but not for minP, coupled with an effect of rTMS
over DLPFC for minP but not for MajP would doubly dissociate the MajP and minP
stages and inform on processing in conditional reasoning
Within-trial rTMS in Conditional Reasoning
If DEAD then BREATHING
Major premise
(Consequent)
(1 sec)
DEAD
Minor premise
(1 sec)
BREATHING
Conclusion
…
…
-1000
-500
0