Cognitive Psychology Series Part 2: Attention & Focus

1. Types of Attention

Think of attention as a resource allocation system with different modes of operation. Just as a city's power grid can direct electricity to different districts based on demand, your brain allocates attentional resources in different configurations depending on the task at hand.

1.1 Selective Attention

Selective attention is the ability to focus on one source of information while filtering out others. It's the "spotlight" or "zoom lens" of attention — you can narrow it to a single word on a page or broaden it to take in an entire scene.

Analogy: Imagine you're at a crowded restaurant. Selective attention is like pointing a directional microphone at your dining companion — their voice comes through clearly while the background noise fades to a murmur. But if someone at the next table mentions your name, your attention snaps toward them involuntarily.

1.2 Sustained Attention (Vigilance)

Sustained attention is the ability to maintain focus on a task or stimulus over an extended period. It's the cognitive equivalent of a marathon — the longer you go, the harder it gets.

Norman Mackworth's Clock Test (1948), developed for the British military, first revealed the vigilance decrement: performance on sustained monitoring tasks drops significantly after about 15-20 minutes. This finding has profound implications for any profession requiring prolonged monitoring — from air traffic control to medical imaging.

1.3 Divided Attention

Divided attention refers to the ability to process multiple sources of information or perform multiple tasks simultaneously. Despite our subjective feeling of multitasking, research consistently shows that true simultaneous processing is extremely limited.

The critical factor is task automaticity. You can walk and talk simultaneously because walking is highly automatic for healthy adults. But try to write an email while having a phone conversation — both tasks requiring language processing — and performance on both suffers dramatically.

# Simulating the dual-task interference effect
# When two tasks compete for the same cognitive resources, performance drops

import random
import time

class AttentionSimulator:
    """Simulates single-task vs dual-task performance."""

    def __init__(self):
        self.base_accuracy = 0.95    # Single-task accuracy
        self.base_rt = 400           # Single-task reaction time (ms)

    def single_task(self, task_name, n_trials=20):
        """Perform a single task with full attention."""
        results = []
        for _ in range(n_trials):
            accuracy = random.random() < self.base_accuracy
            rt = self.base_rt + random.gauss(0, 50)
            results.append({'accurate': accuracy, 'rt': max(200, rt)})

        avg_acc = sum(r['accurate'] for r in results) / n_trials
        avg_rt = sum(r['rt'] for r in results) / n_trials
        print(f"[Single Task: {task_name}]")
        print(f"  Accuracy: {avg_acc:.1%}  |  Mean RT: {avg_rt:.0f} ms")
        return avg_acc, avg_rt

    def dual_task(self, task_a, task_b, resource_overlap=0.7, n_trials=20):
        """
        Perform two tasks simultaneously.
        resource_overlap: 0 = no interference, 1 = complete interference
        """
        penalty_acc = resource_overlap * 0.30   # Up to 30% accuracy drop
        penalty_rt = resource_overlap * 200      # Up to 200 ms RT increase

        results_a, results_b = [], []
        for _ in range(n_trials):
            acc_a = random.random() < (self.base_accuracy - penalty_acc)
            acc_b = random.random() < (self.base_accuracy - penalty_acc)
            rt_a = self.base_rt + penalty_rt + random.gauss(0, 80)
            rt_b = self.base_rt + penalty_rt + random.gauss(0, 80)
            results_a.append({'accurate': acc_a, 'rt': max(200, rt_a)})
            results_b.append({'accurate': acc_b, 'rt': max(200, rt_b)})

        avg_acc_a = sum(r['accurate'] for r in results_a) / n_trials
        avg_rt_a = sum(r['rt'] for r in results_a) / n_trials
        avg_acc_b = sum(r['accurate'] for r in results_b) / n_trials
        avg_rt_b = sum(r['rt'] for r in results_b) / n_trials

        print(f"[Dual Task: {task_a} + {task_b}] (Overlap: {resource_overlap:.0%})")
        print(f"  {task_a} -> Accuracy: {avg_acc_a:.1%}  |  Mean RT: {avg_rt_a:.0f} ms")
        print(f"  {task_b} -> Accuracy: {avg_acc_b:.1%}  |  Mean RT: {avg_rt_b:.0f} ms")
        return (avg_acc_a, avg_rt_a), (avg_acc_b, avg_rt_b)

# Demonstration
sim = AttentionSimulator()

print("=" * 60)
print("EXPERIMENT: Dual-Task Interference")
print("=" * 60)

# Single tasks
sim.single_task("Visual Search")
sim.single_task("Auditory Monitoring")
print()

# Dual task: low overlap (visual + auditory = different modalities)
sim.dual_task("Visual Search", "Auditory Monitoring", resource_overlap=0.3)
print()

# Dual task: high overlap (two verbal tasks = same resources)
sim.dual_task("Reading", "Listening to Speech", resource_overlap=0.8)

1.4 Alternating Attention

Alternating attention is the ability to shift your focus back and forth between tasks that require different cognitive demands. Unlike divided attention (attempting two tasks at once), alternating attention involves sequential switching — fully engaging with one task, then switching to another.

This is what most people actually do when they think they are "multitasking." A student who checks their phone while writing an essay isn't processing both simultaneously — they're rapidly alternating between the two, with a switch cost each time (typically 200-500 ms plus a significant accuracy penalty).

2. Attention Models

The history of attention theory reads like a detective story. Each model proposed a different location for the attentional "bottleneck" — the point where unattended information is filtered out. The debate evolved from early selection (Broadbent) to attenuation (Treisman) to late selection (Deutsch & Deutsch), with each revision prompted by new experimental evidence.

2.1 Broadbent's Filter Model (1958)

Donald Broadbent proposed the first formal model of attention, based on his work with military personnel monitoring multiple radio channels. His Filter Model argues that attention operates as an all-or-nothing filter early in processing:

1. Sensory buffer: All incoming information is briefly held in a sensory store
2. Selective filter: Based on physical characteristics (pitch, location, loudness), one channel is selected
3. Limited capacity channel: Only the selected information receives full semantic processing
4. Unattended information: Completely blocked — no meaning is extracted

Evidence supporting the model: In dichotic listening tasks, participants could accurately report which ear received which message (physical feature) but could not report the content of the unattended message (semantic feature).

Problem: The model couldn't explain why people sometimes notice their own name in the unattended channel (the cocktail party effect) — if unattended information is completely blocked, how does semantic information like your name break through?

2.2 Treisman's Attenuation Model (1964)

Anne Treisman proposed an elegant revision: instead of completely blocking unattended information, the filter attenuates (turns down the volume on) it. Some information, particularly biologically or personally significant stimuli, has a permanently lowered activation threshold and can break through even when attenuated.

2.3 Late Selection Models (Deutsch & Deutsch, 1963)

Deutsch and Deutsch took the most radical position: all incoming information is fully processed for meaning, and selection occurs only at the response/decision stage. The bottleneck is not in perception but in action — you can perceive everything, but you can only respond to one thing at a time.

2.4 Feature Integration Theory (Treisman & Gelade, 1980)

Later in her career, Treisman proposed Feature Integration Theory to explain how attention works in visual search. The theory distinguishes between two stages:

1. Pre-attentive stage: Basic features (color, orientation, size, motion) are processed automatically and in parallel across the entire visual field — no attention required
2. Focused attention stage: Binding features together into coherent objects requires focused attention and occurs serially — one location at a time

# Simulating visual search: Feature search vs Conjunction search
# Feature search is parallel (fast); conjunction search is serial (slow)

import random

class VisualSearchExperiment:
    """Simulates feature vs conjunction visual search tasks."""

    def __init__(self):
        self.base_rt_per_item = 5      # ms per item in feature search
        self.serial_rt_per_item = 40   # ms per item in conjunction search
        self.base_rt = 350             # base reaction time

    def feature_search(self, set_sizes):
        """
        Feature search: target differs in ONE feature (e.g., red among green).
        Reaction time is roughly CONSTANT regardless of set size (parallel).
        """
        print("FEATURE SEARCH: Find the red 'O' among green 'O's")
        print("-" * 50)
        for n in set_sizes:
            noise = random.gauss(0, 15)
            rt = self.base_rt + self.base_rt_per_item * n + noise
            print(f"  Set size {n:3d}: RT = {rt:.0f} ms  (slope ~ {self.base_rt_per_item} ms/item)")

    def conjunction_search(self, set_sizes):
        """
        Conjunction search: target defined by combination of features
        (e.g., red 'O' among red 'X's and green 'O's).
        Reaction time INCREASES linearly with set size (serial).
        """
        print("\nCONJUNCTION SEARCH: Find red 'O' among red 'X' and green 'O'")
        print("-" * 50)
        for n in set_sizes:
            noise = random.gauss(0, 25)
            # On average, search half the items before finding target
            rt = self.base_rt + self.serial_rt_per_item * (n / 2) + noise
            print(f"  Set size {n:3d}: RT = {rt:.0f} ms  (slope ~ {self.serial_rt_per_item} ms/item)")

# Run experiment
exp = VisualSearchExperiment()
sizes = [5, 10, 20, 40]

exp.feature_search(sizes)
exp.conjunction_search(sizes)

print("\nConclusion: Feature search is PARALLEL (flat slope),")
print("Conjunction search is SERIAL (steep slope ~ 20-40 ms/item).")

3. Cognitive Control

While selective attention is about filtering input, cognitive control (or executive attention) is about managing your own behavior — choosing what to focus on, inhibiting inappropriate responses, and flexibly switching between tasks. It's the "CEO" of the attentional system.

3.1 Executive Attention

Executive attention refers to the top-down control processes that regulate where attention is directed. It's mediated primarily by the prefrontal cortex and the anterior cingulate cortex, and it develops slowly through childhood and adolescence — which is why children struggle with impulse control and sustained focus.

Key functions of executive attention:

• Goal maintenance: Keeping your current objective active in working memory
• Conflict monitoring: Detecting when automatic responses conflict with goals (anterior cingulate cortex)
• Attentional set: Configuring the perceptual system to prioritize task-relevant features
• Error detection: Recognizing mistakes and adjusting behavior

3.2 Task Switching

Every time you switch between tasks, you pay a switch cost — a measurable increase in reaction time and decrease in accuracy. This cost arises because switching requires reconfiguring your cognitive "task set" — the goals, rules, and stimulus-response mappings relevant to the new task.

Two components contribute to switch costs:

1. Reconfiguration cost: Time needed to activate the new task set (~200-500 ms)
2. Proactive interference: Residual activation from the previous task set, causing "task-set inertia" that slows performance on the new task

3.3 Inhibitory Control

Inhibitory control is the ability to suppress automatic or prepotent responses when they conflict with your goals. It's essential for self-regulation, social behavior, and flexible responding.

Two key paradigms measure inhibitory control:

4. Limits of Attention

Attention is a limited resource. Understanding these limits is not just theoretically important — it has practical implications for education, workplace design, technology interfaces, and safety-critical systems.

4.1 Cognitive Load Theory (Sweller, 1988)

John Sweller's Cognitive Load Theory is one of the most influential frameworks in educational psychology. It argues that learning is constrained by the limited capacity of working memory, and identifies three types of cognitive load:

4.2 Bottleneck Effects: The Psychological Refractory Period

The Psychological Refractory Period (PRP) demonstrates a fundamental bottleneck in human information processing. When two stimuli requiring responses are presented in quick succession, the response to the second stimulus is delayed — even if the tasks are simple and use different sensory modalities.

Analogy: Think of the PRP as a single-lane bridge. Cars (stimuli) can approach from both directions, but only one can cross at a time. The second car must wait for the first to clear, regardless of how fast each car can travel individually.

Harold Pashler's research (1994) showed that this bottleneck occurs specifically at the response selection stage — deciding what to do — not at perception or motor execution. This suggests a fundamental architectural constraint of the human cognitive system, not just a failure of practice or strategy.

4.3 The Multitasking Myth

Perhaps no topic in attention research has more practical relevance than the question of multitasking. Despite widespread belief that some people are excellent multitaskers, the scientific evidence is clear:

# Modeling cognitive load and its impact on task performance
# Based on Sweller's Cognitive Load Theory

class CognitiveLoadModel:
    """
    Models how different types of cognitive load affect
    learning outcomes and performance.
    """

    WORKING_MEMORY_CAPACITY = 1.0  # Normalized to 100%

    def __init__(self, intrinsic=0.3, extraneous=0.2, germane=0.3):
        self.intrinsic = intrinsic
        self.extraneous = extraneous
        self.germane = germane

    @property
    def total_load(self):
        return self.intrinsic + self.extraneous + self.germane

    @property
    def is_overloaded(self):
        return self.total_load > self.WORKING_MEMORY_CAPACITY

    def learning_effectiveness(self):
        """
        Learning effectiveness depends on germane load
        but only when total load doesn't exceed capacity.
        """
        if self.is_overloaded:
            overflow = self.total_load - self.WORKING_MEMORY_CAPACITY
            # Germane processing is first to be sacrificed
            effective_germane = max(0, self.germane - overflow)
            return effective_germane / self.WORKING_MEMORY_CAPACITY
        return self.germane / self.WORKING_MEMORY_CAPACITY

    def report(self, label=""):
        print(f"\n{'=' * 50}")
        print(f"Cognitive Load Analysis: {label}")
        print(f"{'=' * 50}")
        print(f"  Intrinsic Load:   {self.intrinsic:.0%} (material complexity)")
        print(f"  Extraneous Load:  {self.extraneous:.0%} (poor design)")
        print(f"  Germane Load:     {self.germane:.0%} (learning effort)")
        print(f"  Total Load:       {self.total_load:.0%} / {self.WORKING_MEMORY_CAPACITY:.0%}")
        print(f"  Overloaded:       {'YES' if self.is_overloaded else 'No'}")
        print(f"  Learning Effect:  {self.learning_effectiveness():.0%}")

# Scenario 1: Well-designed lecture
good_design = CognitiveLoadModel(intrinsic=0.35, extraneous=0.10, germane=0.45)
good_design.report("Well-Designed Lecture")

# Scenario 2: Poorly designed lecture (same material)
bad_design = CognitiveLoadModel(intrinsic=0.35, extraneous=0.50, germane=0.45)
bad_design.report("Poorly Designed Lecture (cluttered slides)")

# Scenario 3: Optimized after reducing extraneous load
optimized = CognitiveLoadModel(intrinsic=0.35, extraneous=0.10, germane=0.50)
optimized.report("Optimized Design (max germane load)")

5. Attention & Performance

Understanding attention isn't just about knowing its limits — it's about knowing how to optimize it for peak performance. From Csikszentmihalyi's flow states to evidence-based attention training, this section covers how to harness attention for real-world excellence.

5.1 Flow States (Csikszentmihalyi)

In 1990, psychologist Mihaly Csikszentmihalyi described flow — a state of complete absorption in an activity where attention is so fully engaged that self-consciousness disappears, time distortion occurs, and performance feels effortless. Flow represents the optimal state of attention.

Conditions that promote flow:

• Challenge-skill balance: The task must be challenging enough to require full engagement but not so difficult as to cause anxiety
• Clear goals: You know exactly what you're trying to achieve at each moment
• Immediate feedback: You can tell how well you're doing in real time
• Deep concentration: External distractions are minimized
• Sense of control: You feel capable of meeting the challenge
• Intrinsic motivation: The activity is rewarding in itself

5.2 Vigilance & Sustained Performance

The vigilance decrement — the decline in detection performance over time during sustained monitoring — is one of the most reliable findings in attention research. After 15-20 minutes of monotonous monitoring, detection rates typically drop by 15-50%.

Two theories explain the vigilance decrement:

5.3 Attention Training

Can attention be trained? The evidence is mixed but increasingly promising:

• Meditation: Mindfulness meditation has been shown to improve sustained attention, reduce mind-wandering, and increase cortical thickness in attention-related brain areas (Jha, Krompinger & Baime, 2007)
• Video games: Action video game players show improved attention across multiple paradigms — faster visual search, better spatial attention, and enhanced attentional control (Green & Bavelier, 2003)
• Working memory training: Programs like "N-back training" show limited transfer — improving performance on trained tasks but debatable transfer to untrained tasks (the "far transfer" problem)
• Physical exercise: Acute bouts of exercise temporarily improve attention, and regular exercise is associated with better executive function and reduced ADHD symptoms

6. Advanced Topics

6.1 Neural Networks of Attention

Michael Posner and colleagues identified three functionally and anatomically distinct attention networks in the brain:

Posner's Attention Network Test (ANT) measures all three networks in a single 20-minute computerized task, providing an "attention profile" that has been used extensively in developmental, clinical, and cross-cultural research.

6.2 ADHD & Attention Disorders

Attention Deficit Hyperactivity Disorder (ADHD) affects approximately 5-7% of children and 2.5% of adults worldwide. Despite its name, ADHD is not simply a deficit of attention — it's better understood as a disorder of attention regulation.

People with ADHD can often hyperfocus on highly stimulating activities (video games, novel projects) while struggling to sustain attention on mundane tasks. This pattern is consistent with current understanding of ADHD as involving:

• Dopaminergic dysfunction: Reduced dopamine signaling in the prefrontal cortex and striatum
• Executive function deficits: Difficulty with working memory, inhibitory control, and task switching
• Altered reward processing: Steeper temporal discounting (preference for immediate rewards)
• Default mode network (DMN) interference: Reduced suppression of the DMN during tasks requiring focused attention, leading to mind-wandering

6.3 Eye-Tracking Research

Modern eye-tracking technology has revolutionized attention research by providing a continuous, objective measure of where people direct their visual attention. Since we move our eyes 3-4 times per second (saccades), eye tracking provides a rich temporal record of attentional deployment.

Key findings from eye-tracking research:

• F-pattern reading: On websites, people scan in an F-shaped pattern — reading the first few lines fully, then scanning down the left side (Nielsen, 2006)
• Banner blindness: Users systematically avoid looking at areas that resemble advertisements, even when those areas contain relevant content
• Expert vs novice: Expert radiologists fixate on diagnostically relevant areas within 1-2 seconds, while novices scan broadly and inefficiently
• Emotion and attention: Faces, especially eyes, are powerful attention attractors — a finding leveraged in advertising and user interface design

# Simulating eye-tracking data analysis
# Modeling fixation patterns during visual search

import random

class EyeTrackingAnalyzer:
    """Analyzes simulated eye-tracking data for attention research."""

    def __init__(self):
        self.fixations = []

    def generate_search_data(self, target_loc, n_fixations=15, expertise="novice"):
        """
        Simulate fixation sequence during visual search.
        Experts: fewer fixations, shorter scanpath, faster to target.
        Novices: more fixations, longer scanpath, wandering pattern.
        """
        self.fixations = []
        target_x, target_y = target_loc

        if expertise == "expert":
            # Experts quickly orient toward diagnostic areas
            for i in range(min(n_fixations, 6)):
                progress = (i + 1) / 6
                x = target_x + random.gauss(0, 80 * (1 - progress))
                y = target_y + random.gauss(0, 80 * (1 - progress))
                duration = random.gauss(220, 40)  # Shorter fixations
                self.fixations.append({
                    'x': x, 'y': y,
                    'duration': max(100, duration),
                    'order': i + 1
                })
        else:
            # Novices: broad, unfocused scanning
            for i in range(n_fixations):
                if i < n_fixations - 3:
                    x = random.uniform(0, 1024)
                    y = random.uniform(0, 768)
                else:
                    x = target_x + random.gauss(0, 100)
                    y = target_y + random.gauss(0, 100)
                duration = random.gauss(280, 60)  # Longer fixations
                self.fixations.append({
                    'x': x, 'y': y,
                    'duration': max(100, duration),
                    'order': i + 1
                })

        return self.fixations

    def compute_metrics(self):
        """Compute standard eye-tracking metrics."""
        n = len(self.fixations)
        total_duration = sum(f['duration'] for f in self.fixations)
        avg_duration = total_duration / n if n > 0 else 0

        # Scanpath length (sum of saccade distances)
        scanpath = 0
        for i in range(1, n):
            dx = self.fixations[i]['x'] - self.fixations[i-1]['x']
            dy = self.fixations[i]['y'] - self.fixations[i-1]['y']
            scanpath += (dx**2 + dy**2) ** 0.5

        return {
            'num_fixations': n,
            'total_dwell_time': total_duration,
            'avg_fixation_duration': avg_duration,
            'scanpath_length': scanpath
        }

    def report(self, label):
        metrics = self.compute_metrics()
        print(f"\n{label}")
        print(f"  Fixations: {metrics['num_fixations']}")
        print(f"  Total Dwell: {metrics['total_dwell_time']:.0f} ms")
        print(f"  Avg Fixation: {metrics['avg_fixation_duration']:.0f} ms")
        print(f"  Scanpath: {metrics['scanpath_length']:.0f} px")

# Compare expert vs novice
target = (512, 384)

tracker = EyeTrackingAnalyzer()
tracker.generate_search_data(target, n_fixations=6, expertise="expert")
tracker.report("Expert Radiologist:")

tracker.generate_search_data(target, n_fixations=15, expertise="novice")
tracker.report("Novice Medical Student:")

Exercises & Self-Assessment

Conclusion & Next Steps

In this second chapter of our Cognitive Psychology Series, we've explored the remarkable mechanisms that determine what enters conscious awareness and how effectively we process information. Here are the key takeaways:

• Attention comes in multiple forms — selective, sustained, divided, and alternating — each with different characteristics and limitations
• The bottleneck debate (Broadbent vs Treisman vs Deutsch & Deutsch) reveals that some unattended information is processed, but the degree depends on its significance and current task demands
• Cognitive control — executive attention, task switching, and inhibitory control — is mediated by the prefrontal cortex and develops slowly through adolescence
• Cognitive Load Theory provides a practical framework for designing effective learning experiences by minimizing extraneous load and maximizing germane load
• True multitasking is largely a myth — what we call multitasking is usually rapid alternating, with significant performance costs
• Flow states represent optimal attention — achievable when challenge matches skill and goals are clear
• Three neural networks (alerting, orienting, executive) support different aspects of attention, each with distinct neurotransmitter systems