Cognitive Engineering PSYC 530 Vigilance

1. CognitiveEngineeringPSYC 530Vigilance Raja Parasuraman

2. Overview • Historical Background • Theories of Vigilance • Signal Detection Theory and Vigilance • Enhancing Vigilance in Real World Systems

3. Historical Background

4. Origins of Vigilance Research (1940s) • German submarines sinking Allied naval and cargo ships in English Channel and Bay of Biscay • Royal Air Force (RAF) reconnaissance and anti-submarine warfare flights • Radar and sonar operators on board watching or listening for U-boat contacts

5. Origins of Vigilance Research (Contd.) • Low probability of contact —most missions ended without locating a German submarine • When a contact was missed, most often towards the end of a flight • False reports--leading to bombing of friendly vessels or large marine life

6. Norman Mackworth’s Studies • Field observations of airborne radar and sonar operators • Conducted simulation experiments at MRC Applied Psychology Unit (he was first Director of the Unit) with the Clock Test • Did first laboratory experiments demonstrating the vigilance decrement phenomenon

7. Double Jump = Signal Single Jumps = Neutral events Signals occur unpredictably and infrequently (12 every 30 minutes) Mackworth Clock Test

8. Results (Mackworth, 1948) “Alerted Detection” Subjects: RAF and Naval personnel (25 each) Clock Test: 12 signals/30 min Following training, 2 hour test 100 90 Vigilance Decrement 80 Proportion of Signals Detected (%) 70 60 50 30 60 90 120 minutes Time on Task

9. Characteristics of Vigilance • Low probability of occurrence of signal (< 5%) • Frequent non-signal (noise) or neutral events • Temporal uncertainty of signal (minutes to days) • Prolonged period of performance (30 min - 1 hr) • Need to consider both • vigilance decrement and • overall vigilance level Vigilance Decrement Yes No High Vigilance Level time Low Minimum acceptable level of performance

10. Types of Vigilance Task • Low vs. High Event Rate (Number of non-signal events per unit time) • Simultaneous vs. Successive (Memory load imposed in successive tasks) • Sensory vs. Cognitive (Signal defined on physical vs. symbolic or semantic level)

11. Typical Vigilance Results Reaction Time Detection Rate 100 1000 80 800 60 600 Proportion of Signals Detected (%) Mean Reaction Time for Detected Signals (ms) 40 400 20 200 0 0 10 20 30 40 minutes 10 20 30 40 minutes Time on Task Time on Task

12. Donald Broadbent, MRC Applied Psychology Unit (deceased) Roy Davies, Aston University (retired) Peter Hancock, University of Central Florida Early photo of some guy at George Mason University Joel Warm, University of Cincinnati Leading Vigilance Researchers Norman Mackworth, MRC Applied Psychology Unit (deceased)

13. Theories of Vigilance

14. Theories of Vigilance • Arousal Theory • Habituation Theory • Expectancy Theory • Attentional Resource Theory • Multi-Component Theories

15. Arousal Theory • Decline in detection rate over time (vigilance decrement) is due to a decrease in physiological arousal due to the repetitive nature of the task • Ascending Reticular Activating System — Brain stem activates diverse regions of cortex

16. Arousal Theory: Pros • Behavioral (activity measures), physiological (EEG theta, skin conductance, etc.), and subjective report measures all point to decreased arousal with time on task • Overall level of vigilance is significantly affected by factors that influence arousal (: Caffeine, other stimulant drugs; morning vs. late night performance; varied auditory stimulation, etc. : Alcohol, other depressant drugs; sleep deprivation; loud noise).

17. Arousal Theory: Cons Detection performance Arousal decline: task performance Arousal decline: no task imperative • Decline in arousal correlates with vigilance decrement, but causal relation difficult to show and opposite causal relation is possible(vigilance decrement  arousal decline) • Many factors that change arousal level do not affect the vigilance decrement over time. Stimulant drug: actual Control predicted Control Depressant drug

18. Expectancy Theory: Pros • Subjects perceive and expect signals to occur at a very low rate (Baker, 1961). As a result, they become more “conservative” in responding (in an SDT sense), and adjust their criterion upwards • Leads to a greater likelihood of missing a signal • This further reduces expectancy for a signal, and so on in a “vicious cycle” (Broadbent, 1971) • Support for theory in Reaction Time (RT) and SDT (ß) measures • Support for theory when signal probability is lowered, detection rate decreases and vigilance decrement increases Low expectancy for signal Upward criterion adjustment Further reduction of expectancy for signal Missed signals

19. Expectancy Theory: Cons • Vigilance decrement occurs even when subjects are specifically trained to match their “expectancy” to the actual signal rate (Baddeley & Colquhoun, 1964) • No independent measure of the construct of “expectancy” • The theory cannot account for sensitivity decrement over time in vigilance

20. SDT and Vigilance

21. Hit and False Alarm Rates in Vigilance Tasks Detection Rate False Alarm Rate 100 10 80 8 60 6 Proportion of Signals Detected (%) Proportion of Non-Signals Responded To (%) 40 4 20 2 0 0 10 20 30 40 minutes 10 20 30 40 minutes Time on Task Time on Task • Early vigilance studies did not report FA rates • If they did, they treated them as “guesses” or computed biased measures of accuracy

22. Two Possible Sources for the Vigilance Decrement 1.0 A. Criterion Shift. The vigilance decrement in hit rate is due to an increase in the criterion ( ß), resulting in “downward movement” along a single ROC Increasing Criterion, Fixed Sensitivity time P(H) 0 P(FA) 1.0 1.0 B. Sensitivity Decrement. The vigilance decrement in hit rate is due to a reduction in d’, represented by points on ROCs with lower sensitivity time P(H) Decreasing Sensitivity 0 P(FA) 1.0

23. Sources of the Vigilance Decrement A. Criterion Shift d’ ß 100 Vigilance Decrement in Hit Rate 80 60 Proportion of Signals Detected (%) or 40 B. Sensitivity Decrement 20 0 10 20 30 40 minutes Time on Task ß d’

24. Sensitivity Decrement (Parasuraman, 1979) • High event rate (> 24/minute) • Successive vs. simultaneous task • Dual-task probe RT task suggests that sensitivity decrement is due to depletion of attentional resources with time on task

25. Using Transcranial Doppler Sonography (TCD) to Measure Attentional Resources

26. TCDS Study of Cerebral Hemovelocity in an Air Traffic Control Vigilance Task Target Cue Provided Just Before Critical Event: 100% Valid 70% Valid No Cue

27. 1.05 1.00 .95 .90 0.85 0.8 0 0 10 20 30 Hitchcock et al. (2000) 100% Valid Cue 70% Valid Cue No Cue No Task Control Cerebral Hemovelocity Score 40 Time on Task (minutes)

28. Multi-Component Theory: Arousal, Attentional Resources, and Expectancy (Parasuraman, 1985) • Arousal controls the overall level of vigilance, but not the decrement • The vigilance decrement can be due to either a sensitivity decrement (d’ decrement) or a criterion shift with sensitivity remaining stable (ß increase) • Sensitivity decrement is due to depletion of attentional resources over time. Sensitivity decrement is greater for: • high event rate (observation load) • successive discrimination (memory load) • low signal salience (perceptual load) • Criterion shift is due to expectancy

29. Enhancing Vigilance in Real Systems

30. Air Traffic Control Signal: Two aircraft in “conflict” (within 1 mile vertically or 5,000 ft horizontally)

31. Industrial Inspection and Quality Control Signal: Manufactured product with specified flaw

32. Techniques to Combat Loss of Vigilance • Show signal examples (reduce memory load) • Increase signal salience (reduce perceptual load) • Blinking • Adding coherent motion • Alternate sensory modalities • Lower event rate (if possible) • Insert false signals (TSA program for baggage screeners) • Allow reporting with multiple levels of confidence