Mapping the Journey Through Death: How We Analyzed 5,000 Near-Death Experiences

What happens when we die? It's humanity's oldest question. While we can't answer it definitively, we can study the experiences of those who've come close—and returned to tell the tale.

This post documents our methodology for creating the NDE Journey Flow visualization, which maps the temporal sequence of events in near-death experiences. We'll cover the data, the AI-powered extraction process, the limitations, and what the patterns reveal.

The Data Source

Our analysis draws from the Near Death Experience Research Foundation (NDERF), one of the largest repositories of firsthand NDE accounts in the world.

Database Overview

Each experience includes:

• Narrative: The experiencer's own words describing what happened
• Questionnaire responses: Standardized questions about specific elements
• Greyson Scale score: A validated measure of NDE depth (0–32)
• Demographic data: Age, gender, cause of near-death state

Why NDERF?

1. Volume: Thousands of accounts provide statistical power
2. Consistency: Standardized questionnaire ensures comparable data
3. Firsthand accounts: Unfiltered descriptions in experiencers' own words
4. Longitudinal: 25+ years of collection reveals stable patterns

The Challenge: Extracting Temporal Sequences

Here's the problem: NDE narratives are free-form text. Someone might write:

"I found myself floating above my body. The next thing I knew, I was in this incredibly bright light. My grandmother was there—she'd passed five years earlier. She told me it wasn't my time..."

To create a flow visualization, we need to extract the ordered sequence of events:

1. out_of_body
2. bright_light  
3. deceased_relatives
4. forced_return

Doing this manually for 5,000 experiences would take months. So we turned to AI.

AI-Powered Sequence Extraction

We used OpenAI's GPT-4o-mini model to extract temporal sequences from each narrative. Here's how it works:

The Prompt Engineering

Our prompt instructs the model to:

1. Identify the FIRST thing that happened (valid starts: out-of-body, void/darkness, tunnel, bright light)
2. Extract the sequence of elements in chronological order
3. Identify HOW they returned (choice to return, forced return, or hit a boundary)
4. Provide supporting quotes from the narrative for each element

CRITICAL RULES:
1. The sequence MUST start with the FIRST thing that happened
2. The sequence MUST end with HOW THEY RETURNED
3. Extract at least 3-5 key elements in between
4. If unclear, mark as invalid

Element Vocabulary

We defined 17 standardized elements based on NDE research literature:

Validation Pipeline

Not every extraction is valid. We apply strict post-processing:

1. Must start with valid element: Rejects sequences starting with "life_review" (that's never first)
2. Must end with return element: Every NDE ends with coming back somehow
3. Minimum 3 elements: Sequences too short lack meaningful structure
4. Known elements only: Rejects hallucinated element names

Extraction Results

Why 56% invalid? Several reasons:

• Narrative too vague ("I saw a light and felt peaceful")
• Not a classic NDE structure (shared death experiences, dreams)
• Missing return element (some accounts end mid-experience)
• Narrative focuses on aftermath, not the experience itself

From Sequences to Flows

Once we have individual sequences, we aggregate them into transition probabilities.

Building the Transition Matrix

For each pair of adjacent elements, we count how many times that transition occurs:

-- Simplified version of our transition computation
SELECT 
  s1.element AS from_element,
  s2.element AS to_element,
  COUNT(*) AS transition_count
FROM sequences s1
JOIN sequences s2 
  ON s1.experience_id = s2.experience_id
  AND s2.sequence_order = s1.sequence_order + 1
GROUP BY s1.element, s2.element;

Probability Calculation

We convert counts to probabilities:

P(bright_light | out_of_body) = 
  count(out_of_body → bright_light) / count(out_of_body → anything)

This tells us: "Of experiences that include out-of-body, what percentage go to bright light next?"

Key Findings

1. The Tunnel is a Myth (Sort Of)

Pop culture's favorite NDE element—the tunnel of light—appears in only 22% of experiences. The majority (78%) skip directly from out-of-body or darkness to the light.

Implication: The "tunnel" may be a cultural overlay, not a universal experience.

2. Most Don't Choose to Return

Our data shows:

• 65% are told to return ("It's not your time")
• 25% choose to return (often for family)
• 10% hit a boundary and simply return

This challenges the narrative that experiencers "choose" to come back.

3. Out-of-Body is the Most Common Start

80% of NDEs begin with leaving the body, not with a tunnel or light. This aligns with clinical observations that OBE often precedes other elements.

4. There's No Single "Correct" Path

While patterns exist, individual experiences vary wildly. Some go:

out_of_body → light → beings → realm → return

Others:

void → tunnel → deceased relatives → return

The diversity suggests NDEs aren't a scripted movie—they're deeply personal.

Limitations & Caveats

Selection Bias

• NDERF attracts people motivated to share profound experiences
• Negative or confusing NDEs may be underreported
• Western, English-speaking experiencers are overrepresented

Extraction Limitations

• AI may misinterpret ambiguous narratives
• Some elements co-occur rather than sequence (we treat them as sequential)
• "Forced return" vs "choice to return" can be subjective

Temporal Uncertainty

Many experiencers report that time didn't exist or was meaningless. Imposing a linear sequence may distort experiences that were simultaneous or non-linear.

Survivorship Bias

By definition, we can only study people who returned. The experience of those who didn't return—if there is one—remains unknown.

Reproducibility

All data and code are available for verification:

Data Structure

Each sequence links back to its source:

interface SequenceElement {
  experience_id: string;      // Links to source experience
  element: string;            // The NDE element
  sequence_order: number;     // Position in sequence
  narrative_excerpt: string;  // Supporting quote
  confidence: number;         // AI confidence (0-1)
}

Verification

You can verify any extraction by:

1. Finding the experience by ID
2. Reading the original narrative
3. Checking if the extracted elements match

We store the supporting quote for each element, making verification straightforward.

What's Next?

This analysis opens several research directions:

1. Cultural comparison: Do elements differ across cultures/languages?
2. Temporal analysis: Have NDE patterns changed over 25 years?
3. Correlation with depth: Do deeper NDEs (higher Greyson scores) have more elements?
4. Cause of death: Do cardiac arrest NDEs differ from accident NDEs?

Try It Yourself

Explore the interactive visualization at noeticmap.com/journey. You can:

• Switch between different "story" views
• Hover over flows to see exact numbers
• Click through to read source experiences

---

This research is part of Noeticmap's mission to apply rigorous data science to consciousness research. We believe that by treating these profound experiences with scientific seriousness, we can learn something about the nature of human consciousness—and perhaps, what lies beyond.

---

Appendix: Technical Details

Model Configuration

{
  model: "gpt-4o-mini",
  temperature: 0.2,           // Low for consistency
  response_format: "json",    // Structured output
  max_tokens: 2000
}

Cost Breakdown

Performance

• Processing rate: ~10 experiences/second
• Total extraction time: ~15 minutes for 1,400 experiences
• Database storage: ~50KB per 100 sequences

---

Questions about our methodology? Reach out at research@noeticmap.com