-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
145 lines (128 loc) · 9.58 KB
/
Copy pathindex.html
File metadata and controls
145 lines (128 loc) · 9.58 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<meta name="description" content="An interactive scrolltelling presentation of the 2018 VR Learning Dissertation results and 2026 retrospective on presence and cognitive load.">
<meta property="og:title" content="Are Virtual Simulations Effective?">
<meta property="og:description" content="An interactive data story exploring VR, learning, and a surprising statistical fallacy — built from a 2018 PhD study (n=108).">
<meta property="og:image" content="https://savvides.github.io/dissertation/docs/social-preview.png">
<meta property="og:type" content="website">
<meta name="twitter:card" content="summary_large_image">
<title>Are Virtual Simulations Effective? — Dissertation Results</title>
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500&family=Libre+Baskerville:ital,wght@0,400;1,400&family=Playfair+Display:wght@400;700&display=swap" rel="stylesheet">
<link rel="stylesheet" href="style.css">
<script src="https://d3js.org/d3.v7.min.js" defer></script>
</head>
<body>
<div id="progress-bar"></div>
<header id="hero">
<h1>Teaching Science Lab Safety:<br>Are Virtual Simulations Effective?</h1>
<p class="hook">108 participants. Three conditions. One surprising fallacy.</p>
<p class="author">Philippos Savvides · Arizona State University · 2018</p>
<div class="scroll-indicator" aria-hidden="true">
<span>Scroll</span>
<svg width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
<path d="M7 13l5 5 5-5M7 7l5 5 5-5"/>
</svg>
</div>
</header>
<main id="scroll-container">
<div id="text-column">
<div class="step" data-step="1">
<span class="step-label">01 — The Experiment</span>
<h2>108 Participants, Three Worlds</h2>
<p>What happens when you teach lab safety inside a virtual world instead of a classroom? To find out, 108 university students were randomly assigned to one of three conditions: a fully immersive VR headset, a desktop-based 3D simulation, or a traditional video-and-text control group. Each participant completed the same lab safety curriculum — only the level of immersion changed.</p>
<p>Would more immersion mean more learning?</p>
</div>
<div class="step" data-step="2">
<span class="step-label">02 — The Promise</span>
<h2>Simulations Work</h2>
<p>The initial results were encouraging. Both the VR and Desktop groups significantly outperformed the control group in procedural knowledge and knowledge transfer. Simulations — regardless of immersion level — proved to be an effective teaching tool for lab safety.</p>
<p>But here's the catch: the VR group did not significantly outperform the Desktop group. Higher immersion didn't translate to higher learning gains. Why not?</p>
</div>
<div class="step" data-step="3">
<span class="step-label">03 — The Deeper Question</span>
<h2>But Why?</h2>
<p>The original hypothesis centered on <em>presence</em> — the psychological sense of "being there" inside a virtual environment. The theory was elegant: higher immersion induces stronger presence, and stronger presence drives deeper learning.</p>
<p>When we plot every participant's presence score against their learning gains, the data seems to support this. A clear positive trend emerges across all 108 participants. Case closed?</p>
</div>
<div class="step" data-step="4">
<span class="step-label">04 — The Twist</span>
<h2>Look Closer</h2>
<p>Not quite. A 2026 reanalysis revealed something the original study missed: an <em>ecological fallacy</em>. The overall correlation between presence and learning was real — but it was an artifact of the groups, not the individuals.</p>
<p>When we color each dot by its experimental condition, the picture changes dramatically. Within each group, the relationship between presence and learning virtually disappears. The "trend" was created by the fact that VR participants had both higher presence <em>and</em> higher learning — but one wasn't causing the other.</p>
</div>
<div class="step" data-step="5">
<span class="step-label">05 — The Hidden Cost</span>
<h2>The Price of Novelty</h2>
<p>If presence wasn't the mechanism, what was holding VR back from outperforming desktop? The likely culprit: <em>cognitive load</em>. The VR environment was novel and complex. Participants spent mental energy simply navigating the unfamiliar interface — energy that could have been directed toward learning.</p>
<p>"More immersive" isn't automatically "better for learning" — not until the novelty cost is accounted for.</p>
</div>
</div>
<div id="chart-container">
<!-- D3 visualizations rendered here -->
</div>
</main>
<section id="coda">
<h2>What I Found</h2>
<p>Eight years later, I found a fallacy hiding in my own data. The original 2018 analysis showed that presence correlates with learning gains. The 2026 reanalysis showed the correlation was an artifact of how the groups were sliced — within each condition, it vanishes. The dataset hadn't changed. My eyes had.</p>
<p>If you're designing learning technology, the takeaway is narrower than the theoretical implication. A strong group-level pattern can be seductive and false. Ask whether you're measuring the thing or the bucket. Ask whether the relationship survives at the individual level. It usually doesn't.</p>
<p>If you're specifically building VR for learning, the practical move is to reduce novelty cost before measuring effectiveness — train people on the interface before measuring what they learned with it. The cognitive load of "where do I look, how do I move" steals bandwidth that should be going to the content. Familiarity precedes flow.</p>
</section>
<section id="evidence-lab" aria-labelledby="lab-heading">
<div class="lab-header">
<h2 id="lab-heading">Explore the Evidence</h2>
<p class="lab-intro">Same data, your hands on the controls. Toggle the view between overall and by-condition, switch the y-axis, show or hide the trendlines. The fallacy from step 4 is the easiest to spot — flip from <strong>Overall</strong> to <strong>By Condition</strong> and watch the slope flatten.</p>
</div>
<div class="lab-controls">
<div class="control-group" role="group" aria-label="View mode">
<span class="control-label">View</span>
<div class="control-buttons">
<button type="button" class="lab-toggle active" data-view="overall" aria-pressed="true">Overall</button>
<button type="button" class="lab-toggle" data-view="condition" aria-pressed="false">By Condition</button>
</div>
</div>
<div class="control-group" role="group" aria-label="Trendline visibility">
<span class="control-label">Trendlines</span>
<div class="control-buttons">
<button type="button" class="lab-toggle active" data-regression="on" aria-pressed="true">Show</button>
<button type="button" class="lab-toggle" data-regression="off" aria-pressed="false">Hide</button>
</div>
</div>
<div class="control-group" role="group" aria-label="Y-axis metric">
<span class="control-label">Y-Axis</span>
<div class="control-buttons">
<button type="button" class="lab-toggle active" data-axis="learning" aria-pressed="true">Learning Gains</button>
<button type="button" class="lab-toggle" data-axis="cognitive" aria-pressed="false">Cognitive Load</button>
</div>
</div>
</div>
<div id="lab-chart-container">
<noscript>
<p class="lab-fallback">The Evidence Lab needs JavaScript. The narrative above conveys the same finding without it.</p>
</noscript>
</div>
</section>
<footer id="about">
<div class="about-column">
<h3>How This Was Made</h3>
<p>This site was built entirely through conversation with Claude Code. The full process — from PDF to interactive scrollytelling — is documented in the write-up.</p>
<a href="https://github.com/savvides/dissertation/blob/main/HOW-IT-WAS-MADE.md">Read the write-up →</a>
</div>
<div class="about-column">
<h3>Philippos Savvides</h3>
<a href="https://www.linkedin.com/in/savvides/">LinkedIn →</a>
</div>
<div class="about-column">
<h3>The Data</h3>
<p>The full dataset and original dissertation are open. License: CC BY 4.0 — reuse with attribution.</p>
<a href="data.csv">Download data.csv →</a><br>
<a href="Savvides_asu_0010E_18427.pdf">Dissertation (PDF) →</a>
</div>
</footer>
<script src="script.js"></script>
</body>
</html>