Dual Processes in Reasoning

Minute-Long Mouse-Tracking reveals Conflicting Processes

Eoin Travers
Jonathan Rolison
Aidan Feeney
}

Queen's University Belfast

Response dynamics in reasoning

Typical Results


Explore

Tasks

  • Similarity vs. Categories in Induction
  • Associative vs. Structured knowledge in Induction
  • Descriptions vs. Base-Rate information

Meandering Mice

Dual Processes in Reasoning

  • One or two processes?
  • A less contentious proposition:
    Processes with/without Working Memory
  • Predictions about conflict

Cognitive Reflection Test

An archetypal dual process task?

A bat an a ball together cost €1.10
A ball on its own costs 10c.
How much does the bat cost?

Predictions about Conflict

When should we see it?

Never?

Single-process theories

Selective/Pre-emptive conflict resolution dual processes

One way?

Corrective/Default-Interventionist dual processes

Two ways?

Competitive/Parallel-Competitive dual processes

Intuitive logic

Current Study

Design

  • HTML5 experiment, large computer lab
  • 8x Questions: 4 conflict, 4 control
  • 131 participants(!)
  • Analysis of first 60 seconds mouse data

Conflict

A bat an a ball together cost €1.10
A ball on its own costs 10c.
How much does the bat cost?

Control

A bat and a ball together costs £1.05.
A bat costs £1.
How much does a ball cost?

Results

Choices

Conflict trials Control trials
Heuristic choices 58.3% 15.9%
Correct choices 35.6% 75.1%
Other choices 6.1% 8.8%

Choices

Conflict trials Control trials
Heuristic choices 58.3% 15.9%
Correct choices 35.6% 75.1%
Other choices 6.1% 8.8%

Choices

Conflict trials Control trials
Heuristic choices 58.3% 9.8%
Correct choices 35.6% 80.8%
Other choices 6.1% 9.2%

Summary stats

Comparing responses

Choice N RT Path length Proximity
Correct 181 27.27 8.68 0.99
Heuristic 297 21.02 7.52 1.12
p < .05 < .1 < .05

Baseline condition

Condition N RT Path length
Conflict
(Correct choices)
181 27.27 8.68
Baseline 382 23.75 8.05
Conflict
(Heuristic choices)
297 21.02 7.52
p >.1 >.1

Time course


Quarter speed

Change Points

$P(Heuristic) - P(Other)$
$P(Correct) - P(Other)$

$$ y = \begin{cases} 0 & if\ t\ <\ \tau \\ \beta * t & otherwise \end{cases}\\ \tau \sim Uniform(0, max(t)) $$

Change Points

$P(Heuristic) - P(Other)$
MLE 2.5% 97.5%
$\tau$ 3.829 3.665 3.978
$\beta$ .027 .026 .028
$\sigma$ .012 .011 .013

$P(Correct) - P(Other)$
MLE 2.5% 97.5%
$\tau$ 9.268 9.032 9.484
$\beta$ .014 .014 .015
$\sigma$ .013 .013 .014

Conflict?

Towards heuristic response?

Predictions

No conflict

Single-process theories
Selective dual processes

Conflict

Corrective dual processes
Competitive dual processes
Intuitive logic

Polynomial growth curve

$$ P( Fixation_t) = \\ Logit(\alpha + \beta_{1}t + \beta_{2}t^2\color{red}{^*} + \beta_{3}t^3\color{red}{^*}) \\ \color{red}{^* p < .05} $$

Towards correct response?

Predictions

No conflict

Single process theories
Selective dual processes
Corrective dual processes

Conflict

Competitive dual processes
Intuitive-logic

Polynomial growth curve

$$ P( Fixation_t) =\\ Logit( \alpha \color{red}{^{**}} + \beta_{1}t \color{red}{^{**}} +\\ \beta_{2}t^2 \color{red}{^{***}} + \beta_{3}t^3 \color{red}{^{***}})\\ \color{red}{^{**} p < .01;\ ^{***} p < .001} $$

Unexpected bonus

Early mouse movements predict choices

$$ ^{.} p < .1;\ ^{*} p < .05;\ ^{**} p < .01;\ ^{***} p < .001 $$

Discussion

Theory

Attraction to heuristic response when selecting correct one

...but not the other way around.

Consistent with Corrective, or Default-Interventionist dual process theories

  • Not single process theories?
  • Not selective dual process?
  • Not parallel dual process?
  • Not intuitive logic?

Meandering Mice

Development of decisions over long time scale

Meandering Mice

Questions

  • Is this just eye-tracking?
  • Does this work for other tasks?
  • What's next?

More meandering?

Thanks!