Riassunto esame Human factors and ergonomics, Prof. Di Nocera Francesco, libro consigliato Engineering psychology and human performance. V edition, Wickens

Chapter 3: Signal Detection and Absolute Judgment

1. Overview

Human perception and decision-making often occur under uncertainty — for example,

detecting a weapon in an X-ray scan or identifying a tumor in a radiograph.

This chapter introduces Signal Detection Theory (SDT), which explains how people decide

whether a signal is present or absent amid background noise.

Information must enter through the operator’s senses and be organized and recognized

accurately to ensure correct communication of the displayed information. Thus, an

understanding of how people sense and perceive is essential for display design.

It also explores absolute judgment and multidimensional identification, which extend

these ideas to more complex perceptual decisions.

Choice and preference

Our preferences come from our learning

Alternatives

Each alternative may require a specific behavior and provide reinforcement at a particular rate

and amount.

Paradigm

In the laboratory, choice and preference are studied by employing concurrent reinforcement

programs.

Matching law

When 90 percent of the total reinforcement is provided by schedule A (and 10 percent by

schedule B) about 90 percent of the pigeon pecking occurs in condition A.

This correspondence between relative rate of reinforcement and relative rate of response is

known as the matching law.

Correlation

Each value on the x-axis (relative rates of reinforcement) perfectly

predicts the value on the y-axis (relative rates of behavior).

But the reality is more complex...

DiPerent types of reinforcers and schedules can be active at the

same time. Not all reinforcers have the same perceived value or

ePectiveness for the individual. Some reinforcers may be weighted diPerently according to

their strength or salience. In many situations, behaviors are not controlled by a single source

of reinforcement but by a combination of diPerent schedules.

Y = a + bX

The regression equation tries to find the line

that best fits the points represented by the

data. The regression line, allows us to:

• predict values: if we know X, we can

estimate Y;

• quantify the relationship: the equation

tells us how much Y varies on average

as X varies:

• evaluate the strength of the

relationship: through some statistical indices, we can understand whether the

relationship between the two variables is strong or weak.

• We can make prediction; the best line represents all the points.

Three deviations from correspondence

• Undermatching: the proportions of responses are less extreme than predicted by law.

Undermatching can occur if subjects switch from one response option to another too

often, a tendency that can be reinforced by reinforcements that occur immediately

after the change.

• Overmatching: is the opposite of undermatching and is less common. In this case, the

response proportions of subjects are more extreme than the reinforcement

proportions. Overmatching can occur if there is a switching penalty.

• Bias: occurs when subjects spend more time on an alternative than predicted by the

matching equation. This can occur if a subject prefers a particular environment, lab

area, or response method.

What changes?

GML introduces two important parameters to explain deviations from perfect matching:

• Bias (b): Describes an inherent preference for one

response over another, regardless of the frequency

of reinforcement. For example, an individual might

prefer a certain type of activity because he or she

finds it more familiar or more enjoyable.

• Sensitivity (s): Indicates how precisely the

behavior fits the ratio of available reinforcers.

Lower sensitivity could result from factors such as

distractions or perceived costs in responding to a

source of reinforcement.

Queen’s gambit

As the relative number of games won in which white oPered the queen’s gambit increades, so

did the relative number of games in which white oPered the queen’s gambit. (if you have a

successful strategy you increase)

Education and classroom management

• Example: In a class, two students can choose between actively participating in the

lesson or distracting themselves with other activities (eg, chatting with classmates).

The matching law predicts that students will invest more time in the activity that oPers

more frequent or higher quality reinforcement.

• Application: The teacher can increase positive reinforcement (praise, bonus points,

attention) for desired behaviors such as active participation while reducing the value of

reinforcements for less desired behaviors (ignoring distractions or interrupting in a

controlled manner).

The Matching Law can be applied to design to influence users choices and optimize their

interactions with products, interfaces, and environments. This principle can help designers

create systems that encourage desired behaviors by increasing the perceived value of

reinforcers associated with certain elements.

User interface dosign (UI/UX)

• Example A tune management application oPers several functions (such as setting

eminders or analyzing time spent). The Matching Law predicts that users will ocus on

functions that oPer the greatest perceived value (ease of use, Immediate results).

Application

• Make priority tasks visible and easy to complete.

• OPer immediate positive feedback (completion notifications, success badges) for the

most important tasks.

• Design a simple interface to reduce the cost of interacting with useful functions.

Gamification and engagement

Example: A language learning app like Duolingo uses points, medals, and visual progress to

incentivize regular use. The Matching Law requires users to return to the app regularly if

reinforcement is frequent and rewarding.

Enforcement:

• OPer tangible rewards (points, badges, leaderboards) for completing lessons.

• Show progress in real time to motivate the user.

• Use variable reward mechanisms (eg, random prizes) to keep interest high.

Sensory systems

All sensory systems extract information about four characteristics of the stimulation:

• the sensory modalities and submodalities (e.g., touch as opposed to pain)

• the stimulus intensity o the duration of the stimulation

• its location.

Each system has receptors that are sensitive to some aspect of the physical environment.

These receptors are responsible for sensory transduction, or the conversion of physical

stimulus energy into electrochemical energy in the nervous system.

• The study of sensation and perception involves not only the anatomy and physiology of

the sensory systems but also behavioral measures of perception.

• Psychophysical data obtained from tasks in which observers are asked to detect,

discriminate, rate, or recognize stimuli provide information about how the properties of

the sensory systems relate to what is perceived and acted on.

Vision

Approximately 30% of the cortical surface of a human brain is devoted to representing and

processing information that is mainly visual, making vision the most crucial and highly

developed sense.

Psychophysics

Estimate sensitivity to detect either the presence of some

stimulation or diPerences between stimuli.

The classical methods are based on the concept of a

threshold.

An absolute threshold represents the minimum amount of

stimulation necessary for an observer to tell that a stimulus

was presented on a trial.

A diPerence threshold represents the minimal amount of

diPerence in stimulation along some dimension required to

tell that a comparison stimulus diPers from a standard

stimulus.

Threshold measures confound perceptual sensitivity which they are intended to measure with

response criterion or bias which they are not intended to measure.

2. Signal Detection Theory (SDT)

2.1 Basic Concept

SDT describes how observers discriminate between

• two possible states:

Signal present (signal + noise)

o Signal absent (noise only)

o

The observer’s task is to decide which state exists,

• producing one of four possible outcomes:

Real-world examples:

Detecting a tumor on a medical image (radiology),

• Spotting a hazard while driving,

• Identifying a security threat or system failure.

•

2.2 Probabilities and Measures

From the 2×2 table, we calculate key probabilities:

P(Hit) = hits / (hits + misses)

• P(False Alarm) = false alarms / (false alarms + correct rejections)

• Positive Predictive Value (PPV) = hits / (hits + false alarms)

• Negative Predictive Value (NPV) = correct rejections / (correct rejections + misses)

•

These help quantify how accurate the observer or detection system is.

2.3 The Evidence Variable (X)

Every detection is based on an internal “evidence variable” (X) — the strength of

• sensory evidence for a signal.

Even when no signal exists, random noise can cause X to vary.

• A decision criterion (Xc) divides responses:

• If X > Xc → say “Yes” (signal detected)

o If X < Xc → say “No” (signal absent)

o

Because noise fluctuates randomly:

Sometimes noise alone crosses Xc → False Alarm

• Sometimes the signal doesn’t reach Xc → Miss

•

The greater the separation between signal and noise distributions, the easier the detection.

3. Decision Criterion (β) and Optimality

3.1 Liberal vs. Conservative Responding

Liberal (Risky): low Xc, more “Yes” responses → many hits but also many false alarms.

• Conservative: high Xc, fewer “Yes” responses → fewer false alarms but more misses.

•

Real-life examples:

A cautious doctor (high β) may miss rare diseases.

• A nervous security oPicer (low β) may sound many false alarms.

•

3.2 Factors Abecting Criterion Setting

1. Signal Probability:

If signals are frequent, the criterion should be lower (more liberal).

o If signals are rare, the criterion should be higher (more conservative).

o Formula:

o

βopt=P(N)P(S)βopt=P(S)P(N)

2. Payobs (Costs and Rewards):

The optimal β balances the value of correct detections and the costs of

o errors.

If misses are costly (e.g., in safety-critical jobs), β should be low to avoid them.

o If false alarms are costly (e.g., shutting down a power plant unnecessarily), β

o should be high.

3. Human “Sluggish β” Ebect:

In reality, people don’t adjust β optimally.

o Humans are less sensitive to probability and payoP changes, often

o showing probability matching behavior — they try to balance misses and false

alarms even when that’s not optimal.

4. Sensitivity (d′)

Definition:

The ability to distinguish signal from noise, independent of bias.

When the signal and noise distributions are far apart → high sensitivity (high d′).

• When they overlap heavily → low sensitivity (low d′).

• Measured as:

•

d′=Z(Hit Rate)−Z(False Alarm Rate)d′=Z(Hit Rate)−Z(False Alarm Rate)

where Z is the z-score from the normal distribution.

Sensitivity depends on:

The strength of the signal,

• Observer’s sensory abilities,

• Training and experience.

•

5. The ROC Curve (Receiver Operating Characteristic)

The ROC curve plots P(Hit) (y-axis) vs. P(False Alarm) (x-axis).

It shows how changes in criterion (β) aPect performance.

Upward bowing → high sensitivity.

• Diagonal line → chance performance (no discrimination).

• The area under the ROC curve (A′) quantifies sensitivity without assuming normal

• distributions.

Phenomenal reality

Many things exist as a phenomenon

Gestalt

The psychological “whole” has priority and the “parts” are defined by the structure of the

whole, rather than vice versa

The kanizsa triangle is an optical illusion first described by the Italian psychologist Gaetano

Kanizsa in 1955. The kanizsa triangle is known as a subjective or illusory contour illusion.

We can use this knowledge for design things that lead people what we wont

Eye-tracking

Our eye movement depend on the type of the task we must do

Fixation

Is a pause in eye movement over a specific area of the visual field. These pauses are often

extremely short as the eye continuously performs saccades. Period of time during which the

eyes are relatively still and focused on a specific point in the visual field. They represent

moments when the eyes actively gather information from a particular area.

A clarification

Although the term scanpath is often used informally to describe any recording of eye

movements. Noton and stark (1971) specifically state that a scanpath is achieved by forcing

the subject to look directly (foveally) at any feature to which they wish to pay attention.

Foveal vision

One may think that fixations and saccades give us a clear picture of what an individual

perceives (or pay attention), but this is not exactly the case.

Fixations occur in foveal vision, which accounts for nearly half of the visual information sent

to the brain.

Almost all primate eye movements used to reposition the foveal turn out to be combinations

of five basic types: saccadic smooth pursuit vergence vestibular physiological nystagmus

Other movements (adaptation, accommodation) refer to non-positional aspects of eye

movements (e.g., pupil dilation and lens focusing).

Positional eye movements are of primary importance for the topics we are addressing in this

course.

Eye trackers only track what is recorded in an individual's foveal vision. Unfortunately, this

represents only less than 8 percent of our visual field.

Peripheral vision

• Although we cannot detect the details of objects in the parafoveal and peripheral

regions, we can scan a scene and understand the situation while not examining the

details.

• The fact that a subject has not specifically fixed an element does not mean that he or

she is unaware of its presence.

• The clustering of a number of fixations in a particular region may provide more

indication that something has been watched.

Types of eye-trackers (interface scanning area, specification)

Remote (60s, 70S)

The evolution of eye-tracking based on video technologies has given rise to a new generation

of eye-trackers and opened up the possibility of further uses of eye-tracking.

Corneal reflex

The corneal reflection or glint, is a key component of many eye-tracking systems. It is based

on the principle of capturing and analyzing the reflection of light from the cornea, the

transparent front part of the eye.

Infrared light

An infrared light source is used to illuminate the eye. Infrared light is not visible to the human

eye and does not interfere with normal vision.

Video Recording

A camera equipped with an infrared filter is placed to capture images of the eye.

Corneal reflex

When infrared light illuminates the eye, a small part of it reflects oP the cornea. This reflection

appears as a bright spot on the image captured by

the camera.

Tracking

The eye-tracking system analyzes the position of this bright spot relative to other features of

the eye (such as the pupil or other landmarks). By tracking the movement of this reflection

over time, the system can determine the direction and speed of eye movement.

Visual angle

The concept of visual angle is fundamental:

• Ensure that the screen falls within the "tracking range" of our eye-tracker:

• Ensure that the stimuli on the screen subtend exactly n degrees of visual angle: ensure that

the saccade targets appear at +/- n degrees visual angle from the center of the screen.

While ocular data provided by eye trackers are generally reported in screen pixel coordinates,

important metrics such as saccade amplitude and velocity are reported in visual angle

degrees/degrees per second. saccades and fixations are often distinguished by eye velocity,

which is measured by eye trackers in degrees per second.

Many measures of quality of eye-tracking data (and instrumentation specifications), such as

“accuracy” and “precision”, are reported in degrees of visual angle.

For example, the Eyelink 1000 plus is accurate to < 0.5 degrees of visual angle.

Calibration

Before the position of the corneal reflex can be accurately interpreted, the yes racking system

must be calibrated. specific points or following certain patterns while the system records the

corresponding movements of the reflex. This allows the system to establish a mapping

between the position of the reflex and the direction of gaze.

Contemporary eye-trackers are extremely accurate and enable:

• Calibrate participants' eyes in seconds;

• Trace diPerent populations (the shape of the eyes is not always the same;

• Keep calibrations for long periods of time.

Head-tracking

Some eye-tracking systems also incorporate

head-tracking

capabilities to compensate for head movements. By tracking the position of the user's head,

the system can adjust the interpretation of the corneal reflex position to account for changes

in the viewing angle.

Software

• Contemporary eye trackers are equipped with software suites that produce visualizations of

eye data and automate tasks that previously took weeks.

• The output of these software packages helps to highlight where the user looked, how long

they looked, and what scanning pattern they performed.

Heatmap

• A heatmap is a visualization that uses diPerent colors to show the amount of fixations made

by participants or the amount of time that areas are fixed.

• Red is typically used to indicate a relatively high number of fixations or duration, green the

minimum, with varying levels in between.

Area of Interest

An area of interest (AOl) is a specific region or area within a visual scene.

AOls can be derived from scanpath inspection, or defined a priori.

Pupillary diameter

• The diameter of the pupil is regulated mainly by the autonomic nervous system.

• It can be influenced by emotional and cognitive processes.

• Mental workload has been associated with changes in pupil size.

6. Applications of SDT

6.1 Medical Diagnosis

Radiologists detect tumors using SDT principles.

• Checklists and “aids” increase sensitivity by focusing attention.

• Bias shifts with disease prevalence:

• Low prevalence → conservative criterion (high β)

o High prevalence → liberal criterion (low β)

o

U.S. physicians show lower biopsy criteria (more false positives) than U.K. physicians.

•

6.2 Eyewitness Identification

Witnesses must decide if a suspect was the culprit (signal) or not (noise).

• Sequential lineups encourage conservative responses and reduce false

• identifications.

Feedback (“You picked the suspect”) increases confidence even when wrong —

• a false certainty ePect.

Independent administrators are recommended to avoid bias.

•

6.3 Alarm and Alert Systems

Alarms function as detectors with thresholds (β).

• Low β = frequent alerts (many false alarms).

• High β = few alerts (risk of missing true dangers).

• Frequent false alarms lead to the “cry wolf” ebect — users start ignoring alarms.

• Solutions:

• 1. Multi-level alarms (graded urgency),

2. Slightly higher automated thresholds,

3. Keeping humans in the loop,

4. Training users to understand false alarms statistically.

7. Vigilance or sustained attention

7.1 Definition

Vigilance = sustained attention to detect infrequent and unpredictable signals over long

periods. (Vigilance is about attention, attention is not a unitary concept, we have diPerent

meaning of attention. Tasks in which vigilance is important like driving)

Examples:

Air traPic controllers monitoring radar,

• Security inspectors,

• Animal predators waiting for prey.

•

Historical Roots: Mackworth's Clock Test

• 1948 study on radar operators.

• Participants monitored rare double jumps.

• Detection declined after 15-30 minutes.

Our ability to track what is going on is the expression of vigilance.

7.2 Findings

Vigilance decrement: performance drops after ~30 minutes.

• Decrement arises from:

• Decreased sensitivity (fatigue, overload),

o Conservative bias shifts (expecting fewer signals).

o

Characteristic of vigilance tasks

• Low event frequency.

• High importance of detection.

• Minimal feedback.

• Monotonous environment.

7.3 Factors Influencing Vigilance

Weak or infrequent signals → lower sensitivity.

• Uncertain timing or location of events → harder detection.

• High event rate → overload.

• Extra working-memory load → fatigue and distraction.

• Low signal probability → conservative bias (high β).

•

7.4 Theories

1. Arousal Theory (physiological alertness): monotony/low reduces alertness → lower

signal and noise activity/Tasks lack variety. Countermeasure: increase engagement

2. Sustained Demand Theory (fatigue): vigilance tasks require ePort and cause mental

fatigue (resource depletion). sustained attention consumes cognitive resources. Over

time responses slow. Countermeasure: breaks, rotation, automation.

3. Expectancy Theory (probability and readiness): after missing signals, people expect

fewer signals, raising β — a vicious circle. Countermeasure: predictive cues or higher

event rates.

4. Mind-wandering (attention drift): Monotony-thought drift. Control fails to reorient

focus. Countermeasure: Feedback, adaptive pacing.

7.5 Countermeasures

Show examples of targets to reduce memory load,

• Increase signal salience (brightness, contrast, or multimodal cues),

• Reduce event rate,

• Provide rest or automation support,

• Train automatic recognition skills.

•

The challenge of designer

• Humans not built for constant monitoring.

• Fatigue and under-stimulation accumulate.

• Design must compensate for limits.

Real world vigilance failures:

• Air traPic control lapses - Security Oversights - Healthcare monitoring errors

• Vigilance in automation contexts

• Automation removes stimulation but adds responsibility

• Automation

• Psychological

• Costs of Monitoring

• High mental ePort despite few actions.

• Low feedback increase