Cognitive Control during Antisaccade Performance

1
Cognitive Control during Antisaccade Performance

• Sam Hutton
• Department of Psychology
• University of Sussex
• Ben Tatler
• Department of Psychology
• University of Dundee

2
Cognitive Control

• Complex behaviour requires the monitoring of
  ongoing action and subsequent behavioural
  adjustment in order to prevent, detect and (if
  necessary) correct erroneous responses.
• This monitoring is particularly important in
  situations in which execution of the correct
  behavioural response requires the inhibition or
  suppression of an over-learned or habitual
  response.

3
Cohen / Miller / Botvinick model
Active / Working Memory
Goals / Intentions / Representations of
context (DLPFC)
Conflict Monitoring (ACC) CONFLICT

BIAS
I N P U T
Info processing in Posterior Association Cortex
Motor Response
Mutual Inhibition
4
Cohen / Miller / Botvinick model

• Representations of context information in DLPFC
  bias information processing down line.
• Mutual inhibition ensures that biasing the
  correct pathways inhibits processing in the
  incorrect pathways
• A conflict monitoring process (mediated by ACC)
  increases activation in DLPFC, resulting in
  increased bias (control), and a reduction in
  conflict
• One consequence of increased cognitive control is
  a shift to a more conservative response strategy
  (Botvinick et al, 2001) resulting in increased
  correct latencies, and a reduction in the
  probability of error (c.f. Rabbitt, 1966).
• Compelling support from computational modelling,
  but surprisingly little behavioural evidence for
  ongoing adjustments in cognitive control (Mayr
  2003).

5
An alternative model

• The Cohen model is a closed feedback circuit
  there is no role for conscious awareness.
• Dehaene and colleagues (20012003) suggest an
  alternative account.
• The ACC and DLPFC act together as key nodes in a
  conscious neuronal workspace.
• The ACC computes and disseminates information
  concerning the likelihood of reward.
• According to this account, behavioural (and
  neural) indications of cognitive control should
  only occur if conflict is conscious.

6
Cognitive Control and Antisaccades

• The antisaccade task requires participants to
  execute a novel response in place of a highly
  prepotent one.
• Mokkler Fischer (1999) demonstrated that
  participants are unaware of around 50 of the
  errors they make.
• AIMS
• 1) To determine whether the antisaccade task can
  provide evidence in support of ongoing
  adjustments in cognitive control (c.f.
  Neiuwenhuis, 2001)
• 2) To test the the two competing hypotheses
  concerning the role of conscious awareness in
  cognitive control.

7
Repetition Priming and Antisaccades

• Other factors may also influence antisaccade
  performance on a trial by trial basis.
• Salience maps serve to identify locations that
  merit further processing.
• Activation levels reflect bottom up (e.g. size /
  luminance) and top down influences (e.g. strategy
  / expectation).
• Residual effect of activation in these maps
  persists beyond the duration of a single trial
  (e.g. Dorris et al, 2000 Fecteau et al, 2004).
• Mayr et al (2003) conflict adaptation effects
  reflect stimulus specific priming.
• AIM 3) Does stimulus specific priming affect
  antisaccade performance?

8
The analysis

• Needed a lot of trials We used baseline data
  from several psychopharmacology studies /
  undergraduate projects.
• 41 Participants performed a total of 13,798
  antisaccade trials (blocks of 72, 200ms gap,
  targets /- 5 and 10 degs).
• Traditional analysis
• Calculate mean latencies (single error rate) for
  each participant.
• BUT optimal contingency analysis should treat
  each trial as a data point whilst taking
  individual differences into account
• Alternatives
• 1) Randomised Block ANOVA with subjects as a
  random factor
• 2) Mixed Model Analysis (aka multilevel
  modelling).
• This approach does not violate assumption of
  independence and copes with different numbers of
  repeated measurements.

9
General Sample Metrics

• (Traditional analysis)
• Average Error rate 23.3 (SD 15.4)
• Trend for increased errors to targets appearing
  on the right (24.7 vs 21.4, p 0.08)
• Errors more likely to occur to targets closest to
  fixation (29.4 vs 21.2, p lt 0.001)
• Average correct latency 232 msec (SD 38)
• Average error latency 137 msec (SD 28)
• Average latency to correct 131 msec (SD 55)
• 62 of the errors were corrected within 120 ms a
• 33.7 were corrected within 85 ms
• 24.6 of errors were corrected so quickly an
  accurate FEP was achieved faster than the average
  latency for correct trials.

10
Hemifield by Trial Type Interaction for Latencies
Saccades to the right are faster ANOVA on
averages (F(1,40) 6.627, p 0.014) Randomised
Block (F(1,78) 5.5, p lt 0.05) Multilevel
Modelling (IGLS 35, p lt 0.000001)
11
Contingency Effects Error Rates

• Probability of making an error greater after an
  error (contrary to models of cognitive
  control)

Main effect of Previous trial type (F(1,40
4.18, p lt 0.05) (ML analysis also significant)
Borderline interaction between previous trial
type and hemifield congruency F(1,40) 3.15, p
0.08) Following a correct antisaccade,errors less
likely to occur when target is in same hemifield
(19.5 vs 24).
12
Contingency Effects Latencies

• 3-way interaction between current trial outcome,
  previous trial type, and hemifield congruency.
• Normal F(1,32) 5.1, p lt 0.05 RB F(1,90)
  4.8, p lt 0.05

2-Way interaction between current trial outcome
and hemifield congruency after correct
antisaccades Correct latencies faster if
congruent (219 vs 228) but incorrect latencies
faster if incongruent (144 vs 138). E.g.
saccades faster if they are in the same direction
as on previous trial.
13
Contingency Effects Latencies

• Is there any evidence for cognitive control?
• Quickly (?80ms) vs Slowly (?180ms) corrected
  errors

Main effect of previous trial type, p
0.01 Main effect of current trial, p lt 0.001 No
interaction
Post error slowing after slowly corrected
(conscious?) errors Post error speeding after
quickly corrected (unconscious?) errors
14
Summary

• Errors more likely to occur after errors
• Errors least likely to occur after correct
  responses if target appears in same hemifield as
  previous trial.
• No overall effect of previous trial type on
  correct latencies (no post error slowing)
• Following a correct response, correct responses
  are faster if the target appears in the same
  hemifield and error responses are faster if the
  target appears in the opposite hemifield as on
  the previous trial (response priming)
• Post error slowing observed only after errors
  that were corrected after 180 msec (e.g. errors
  which the participant is more likely to be aware
  of).
• Post error speeding observed after errors that
  were corrected within 80 msec (e.g. errors which
  the participant is less likely to be aware of).

15
Conclusions

• In general trial N-1 response priming effects
  have a greater influence on trial N performance
  than trial N-1 outcome.
• These priming effects reflect residual activation
  in salience maps that provide the goal for
  saccades.
• Intention activation fluctuates with a time
  course that extends over several trials.
• There is some evidence for post-error slowing
  after slowly corrected errors (supporting
  Dehaenes model)
• Quickly corrected and slowly corrected errors may
  reflect different processes
• Slowly corrected errors occur when intention
  activation is low
• Quickly corrected errors may occur when
  activation levels are high - and may be adaptive.

16
Conclusions

• The results generally support a parallel race
  model of antisaccade performance.
• However, error latencies are not faster for
  targets appearing close to fixation (but errors
  are much higher), and not all correct responses
  are programmed in parallel with the error.
• Future research will replicate these results in a
  larger sample (N100,000).
• These techniques may also be useful for exploring
  antisaccade deficits in patient populations (e.g.
  schizophrenia, lesions).
• Mixed models a powerful technique for exploring
  contingency effects in the antisaccade paradigm.