File size: 6,776 Bytes
08a5c22 | 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 | ---
license: cc-by-4.0
tags:
- education
- psychometrics
- DIF
- differential-item-functioning
- item-response-theory
- assessment
---
# DIF Dataset Comparison & Collected Data
A curated collection of educational assessment datasets evaluated for **Differential Item Functioning (DIF)** analysis, with actual downloaded data where available.
## What's Here
### 📊 Comparison Spreadsheets
| File | Description |
|------|-------------|
| `DIF_Dataset_Comparison_VERIFIED.xlsx` | Main Excel with 3 sheets: Master Comparison, Column Inventory, DIF Readiness Checklist |
| `DIF_Dataset_Master_Comparison.csv` | Dataset-level comparison (responses, demographics, text, scale, access) |
| `DIF_Dataset_Column_Inventory.csv` | Every verified column per dataset with data types and sources |
| `DIF_Dataset_Readiness_Checklist.csv` | ✅/❌/⚠️ checklist: Can you do MH DIF? Distractor DIF? Text-based DIF? |
### 📁 Actual Data (`data/` folder)
#### `data/ednet/` — EdNet (Riiid/Santa, Korean TOEIC prep)
| File | Description | Size |
|------|-------------|------|
| `kt1_merged_500k.parquet` | 500K student responses merged with question metadata | 9 MB |
| `questions.parquet` | All 13,169 questions (IDs, correct answer, part, tags — NO text) | 323 KB |
| `item_stats.parquet` | Per-item statistics (p-value, n, mean elapsed time) | 397 KB |
| `answer_distributions.parquet` | Distractor analysis: count of each answer option per item | 154 KB |
| `student_stats.parquet` | Per-student accuracy and timing | 74 KB |
**Columns in kt1:** `timestamp, solving_id, question_id, user_answer (a/b/c/d), elapsed_time, subject_id, correct_answer, is_correct`
**DIF status:** ✅ Selected answers, ✅ Correct answer, ❌ No question text, ❌ No demographics
#### `data/assistments/` — ASSISTments 2009 (Massachusetts middle school math)
| File | Description | Size |
|------|-------------|------|
| `assist2009_expanded.parquet` | 274K item-level responses (expanded from nested format) | 608 KB |
| `assist2009_item_stats.parquet` | Per-skill statistics (139 skills) | 9 KB |
| `assist2009_student_stats.parquet` | Per-student accuracy (4,148 students) | 45 KB |
**Columns:** `user_id, item_idx, skill_id, skill_name, correct (0/1), attempt_count, answer_type`
**DIF status:** ❌ This HF version lacks gender. Original 2009 file has gender (~50% missing). **2012-2013 Challenge version** (download from [ASSISTments site](https://sites.google.com/site/assistmentsdata/datasets)) has gender + school + economically_disadvantaged + ELL + IEP.
#### `data/pisa/` — PISA 2018 (schema + download script)
**Too large for HF upload (~1 GB). Use the download script instead.**
- `schema.json` — Complete file/column documentation
- `scripts/download_pisa_locally.py` — Run locally to download + process
**What PISA has:** ✅ Individual scored item responses (0/1/2), ✅ Rich demographics (gender, SES, immigration, country), ❌ No raw selected answer (A/B/C/D), ⚠️ Question text only for ~30-50% of items (in separate PDFs)
#### `data/timss/` — TIMSS 2019 (schema + processing script)
**Requires manual download from IEA website.**
- `schema.json` — File structure documentation
- `scripts/download_timss_locally.py` — Process after manual download
#### `data/foundational_assist/` — FoundationalASSIST (schema only)
**Gated dataset — requires data agreement on HF.**
- `schema.json` — Complete table/column documentation
- Request access: https://huggingface.co/datasets/ASSISTments/FoundationalASSIST
**What it has:** ✅ Full question text (HTML), ✅ Student answer text, ✅ Distractor options, ❌ No demographics
#### `data/oulad/` — OULAD (schema only)
- `schema.json` — Table/column documentation
- Download from: https://analyse.kmi.open.ac.uk/open_dataset (or DOI: 10.1038/sdata.2017.171)
**What it has:** ✅ Rich demographics (gender, IMD/SES, disability, age), ❌ No item-level responses (assessment-level only)
## Quick DIF Feasibility Matrix
| Dataset | Individual Responses | Selected Option | Question Text | Demographics | **Can Do MH DIF?** |
|---------|---------------------|----------------|---------------|-------------|-------------------|
| **EEDI** (your current) | ✅ | ✅ A/B/C/D | ✅ | ✅ Gender, SES | ✅ **YES** |
| **PISA 2018** | ✅ Scored 0/1/2 | ❌ | ⚠️ ~30-50% PDFs | ✅ Gender, SES, Immigration | ✅ **YES** |
| **TIMSS 2019** | ✅ Scored | ❌ | ⚠️ ~30-50% PDFs | ✅ Gender, SES proxy | ✅ **YES** |
| **NAEP** | ❌ public | ❌ | ⚠️ partial | ✅ (restricted only) | ❌ **NO** (restricted) |
| **ASSISTments 2012** | ✅ | ❌ | ❌ | ⚠️ Gender (gaps) | ⚠️ **PARTIAL** |
| **FoundationalASSIST** | ✅ + answer text | ✅ | ✅ | ❌ | ❌ **NO** (no groups) |
| **EdNet** | ✅ | ✅ a/b/c/d | ❌ | ❌ | ❌ **NO** (no groups) |
| **Junyi** | ✅ | ❌ | ❌ | ❌ | ❌ **NO** |
| **OULAD** | ❌ assessment-level | ❌ | ❌ | ✅ | ❌ **NO** (no items) |
## Download URLs (Verified Working)
| Dataset | URL | Registration |
|---------|-----|-------------|
| **PISA 2018 COG** | https://webfs.oecd.org/pisa2018/SPSS_STU_COG.zip (456 MB) | None |
| **PISA 2018 QQQ** | https://webfs.oecd.org/pisa2018/SPSS_STU_QQQ.zip (478 MB) | None |
| **TIMSS 2019** | https://timssandpirls.bc.edu/timss2019/international-database/ | None (website nav) |
| **ASSISTments** | https://sites.google.com/site/assistmentsdata/datasets | None (Google Drive) |
| **FoundationalASSIST** | https://huggingface.co/datasets/ASSISTments/FoundationalASSIST | HF gated |
| **EdNet** | https://huggingface.co/datasets/mgor/EDNet | None |
| **OULAD** | https://analyse.kmi.open.ac.uk/open_dataset | None |
| **NAEP** | https://www.nationsreportcard.gov/nde/ | Restricted-use license |
## ⚠️ Corrections from Common Misconceptions
1. **PISA/TIMSS do NOT give raw selected answers** (A/B/C/D) — only scored 0/1/2
2. **NAEP does NOT provide individual responses publicly** — restricted federal license required
3. **ASSISTments 2015 does NOT have a gender column** — widely misclaimed
4. **Junyi does NOT have gender** — confirmed absent in all versions
5. **OULAD has great demographics but NO item-level responses** — assessment-level only
6. **FoundationalASSIST has the best item content but ZERO demographics**
7. **EdNet has excellent response data but NO demographics at all**
## Recommended Path for DIF Research
1. **PISA 2018** — Best combination of individual item responses + demographics. Run `scripts/download_pisa_locally.py`.
2. **FoundationalASSIST** — Best for distractor-level analysis if you can get demographic linkage (email etrials@assistments.org)
3. **ASSISTments 2012-2013** — Best ASSISTments version for demographics (gender, SES, ELL)
|