documentation/users.md

# Users Dataset Documentation

## Overview

**Dataset Name:** `users.parquet`
**File Pattern:** `users14*.dat`, `questions*.dat`
**Time Period:** 2000-2022
**Data Type:** Survey responses from psi experiment participants

> **🔄 STATUS:** Complete
> **🎯 CONFIDENCE:** 90% - Well-documented with clear processing logic

---

## What This Dataset Contains

This dataset contains survey responses from users who participated in various psi (parapsychology) experiments. The survey collected demographic information, location data, and responses to two types of psychological questionnaires:

1. **Psi questionnaire** - 15 questions about belief in and experiences with psi phenomena
2. **Hemispheric dominance questionnaire** - 10 questions assessing left-brain vs right-brain cognitive preferences

### Real-World Context

Researchers collected this data to explore potential correlations between:
- Belief in psi phenomena and test performance
- Hemispheric brain dominance and psi abilities
- Geographic/demographic factors and psi performance

This dataset serves as the **demographic anchor** for all other experiment datasets and can be joined using `username_hash`.

---

## Data Schema

### Identity & Demographics

| Column | Type | Description | Possible Values |
|--------|------|-------------|-----------------|
| `username` | string | User-chosen username | Any string (removed if `--exclude-pii` flag used) |
| `username_hash` | string | SHA-256 hash of username for joining | 64-character hex string |
| `email` | string | User email address (2014+) | Valid email or NULL (removed if `--exclude-pii` flag used) |
| `timestamp` | datetime | Survey completion timestamp | Timezone-aware datetime |

### Location Information

| Column | Type | Description | Notes |
|--------|------|-------------|-------|
| `city` | string | User's city | Free-text, may contain commas |
| `state` | string | State/province | Free-text |
| `coordinates` | string | Geographic coordinates | Format varies, often "lat,lon" |
| `country` | string | Country | Free-text |

> **⚠️ DATA QUALITY NOTE:** Location fields are user-entered free text with inconsistent formatting and many missing values. Parse with caution.

### Psi Questionnaire (15 questions)

Columns: `psi_01` through `psi_15`

- **Type:** Integer (1-5) or NULL
- **Scale:** 1 = Strongly Disagree, 5 = Strongly Agree
- **Questions cover:** Belief in telepathy, precognition, psychokinesis, personal psi experiences

### Hemispheric Dominance Questionnaire (10 questions)

Columns: `hemi_01` through `hemi_10`

- **Type:** Integer (1-5) or NULL
- **Scale:** 1 = Strongly Disagree, 5 = Strongly Agree
- **Purpose:** Assess left-brain (analytical) vs right-brain (intuitive) cognitive preferences

### Metadata

| Column | Type | Description |
|--------|------|-------------|
| `how_find` | string | How the user found the survey (free text) |
| `na_count_psi_and_hemi` | integer | Count of unanswered questions in psi + hemi surveys |
| `file_type` | string | Source file type: "users14" or "questions" |
| `source_file` | string | Original filename (only if `--audit` flag used) |
| `source_row_number` | integer | Row number in source file (only if `--audit` flag used) |

---

## Data Processing Notes

### Source File Formats

The dataset combines two different file formats:

1. **users14.dat** (2000-2015)
   - Format: `Username, Timestamp, City, State, Coordinates, Country, Psi1-15, Hemi1-10, HowFind, na_count`
   - No email field
   - Simple timestamp format

2. **questions.dat** (2014-2022)
   - Format: `Username, Timestamp, Email, City, State, Coordinates, Country, Psi1-15, Hemi1-10, HowFind`
   - Includes email field
   - Timestamp format: "Wed Jan 1 00:33:44 2014"
   - `na_count` calculated during processing

### Data Cleaning Challenges

> **🚧 OUTSTANDING:** Need to document specific cleaning rules applied to location parsing

**Major challenges addressed:**

1. **Dirty CSV format** - Username and location fields contain commas, requiring custom parsing that works backwards from the clean Psi/Hemi columns (which only contain values 1-5)

2. **Duplicate usernames** - Deduplication rules:
   - Keep row with **fewest NA responses** in Psi and Hemi columns
   - If tied on NAs, keep **oldest timestamp**
   - Cross-file duplicates handled (users14 vs questions)

3. **Mixed date formats** - Two different timestamp formats require format-specific parsing

4. **Test users** - Rows with usernames starting with `_test99` are automatically filtered out

### Known Data Quality Issues

> **🎯 CONFIDENCE:** 75% - Some edge cases in location parsing may not be fully documented

- **NA usernames:** Rows with username "NA" or NULL are **retained** (as of June 18, 2025)
- **Location data:** Highly inconsistent, user-entered free text
- **HowFind field format change:** September 2002 - changed from yes/no to descriptive text

---

## Usage Examples

### Joining with Experiment Data

```python
import pandas as pd

# Load users and experiment data
users = pd.read_parquet('users.parquet')
card_results = pd.read_parquet('card.parquet')

# Join on username_hash
combined = card_results.merge(
    users,
    on='username_hash',
    how='left'
)

# Analyze psi belief vs performance
```

### Filtering Complete Surveys

```python
# Get users who answered all questions
complete_surveys = users[users['na_count_psi_and_hemi'] == 0]
```

---

## Statistical Summary

> **🚧 OUTSTANDING:** Add row counts and coverage statistics after processing

- **Total unique users:** _[To be added]_
- **Duplicates removed:** _[Logged in processing stats]_
- **Date range:** 2000-2022
- **Source files:** users14.dat + questions.dat

---

## Privacy & Ethics

### PII Protection

When using the `--exclude-pii` flag:
- `username` column is removed
- `email` column is removed
- `username_hash` is **retained** for joining with other datasets

### Recommended Usage

For publication or sharing:
```bash
python scripts/process_all.py --exclude-pii
```

This ensures user privacy while maintaining data linkage capabilities.

---

## Related Datasets

This dataset can be joined with:
- **card.parquet** - Basic ESP card test results
- **cardd.parquet** - Card drawing test results
- **cardS.parquet** - Sequential card test results
- **rv.parquet** - Full remote viewing test results
- **rvq.parquet** - Quick remote viewing test results
- **location.parquet** - Coordinate remote viewing results
- **lottery.parquet** - Lottery number prediction results

**Join key:** `username_hash`

---

## References & Context

- Original survey collected demographic and psychological profiles of psi experiment participants
- Survey questions designed to assess belief in psi phenomena and cognitive style
- Data cleaning pipeline handles 20+ years of format variations and dirty user input

---

**Last Updated:** 2025-01-09
**Processor:** `src/processors/users_processor.py`
**Schema Version:** Unified (combines users14 and questions formats)