# Reforming Tertiary Education — Process Tracing Database Design A methodological development repository dedicated to **conceiving, designing, and iterating** on a database schema for performing **abductive process tracing** with **Inference to the Best Explanation (IBE)**. The substantive case is a process of reform in tertiary education; the methodological contribution is the schema itself. > Repositório de desenvolvimento metodológico dedicado a **conceber, projetar e iterar** um esquema de banco de dados para **process tracing abdutivo** com **Inferência à Melhor Explicação (IBE)**. O caso substantivo é um processo de reforma da educação superior; a contribuição metodológica é o esquema em si. --- ## 1. Purpose / Propósito **EN.** This repository documents the *process of thinking through* how to build an evidence database for abductive process tracing. It is not primarily a research data repository — it is a record of methodological reasoning. The core question driving the design is: given that available theoretical explanations are individually *insufficient*, *underspecified*, or *misspecified*, how should evidence be structured so that competing explanations can be compared transparently and any reader can audit exactly where they disagree? The templates represent successive attempts to answer that question. Each version reflects a methodological decision — documented in [`NEWS.md`](NEWS.md) — and the conversations that produced them are preserved as a paper trail for the reasoning behind each iteration. **PT.** Este repositório documenta o *processo de reflexão* sobre como construir um banco de evidências para process tracing abdutivo. Não é primariamente um repositório de dados de pesquisa — é um registro de raciocínio metodológico. A pergunta central é: dado que as explicações teóricas disponíveis são individualmente *insuficientes*, *subespecificadas* ou *malespecificadas*, como a evidência deve ser estruturada para que explicações concorrentes possam ser comparadas com transparência? > **Terminological note.** *Insufficient* = missing scope. *Underspecified* = vague mechanisms. *Misspecified* = premises contradicting empirical reality (following van Fraassen's *empirical inadequacy*). --- ## 2. Methodological Framework / Marco Metodológico The project sits at the intersection of four ideas: 1. **Explaining-outcome process tracing (abductive)** — constructing a comprehensive explanation for a particular complex outcome through iterative movement between evidence and theory. This is not deductive theory-testing; the iterative dialectic is the method itself. 2. **Inference to the Best Explanation (IBE)** — given the assembled evidence, asking which of the candidate explanations performs best in accounting for the observed outcome. 3. **Process tracing with imperfect theories** — explicitly treating theoretical insufficiency, underspecification, and misspecification as a normal starting condition, not a defect. 4. **Bayesian inference with verbal probability scales and decibel weight of evidence** — following Fairfield & Charman (2022), each piece of evidence is evaluated by assigning verbal likelihoods under each rival hypothesis, computing likelihood ratios, and expressing cumulative weight in decibels. No Van Evera tests (Hoop, Smoking Gun, etc.) are used. Full premises and their rationale are documented in [`METHODOLOGY.md`](METHODOLOGY.md), a *living document* updated incrementally. --- ## 3. Why Abductive, Not Deductive / Por Que Abdutivo, Não Dedutivo Standard (deductive) process tracing assumes a well-specified theory with clear observable implications. That premise breaks down when: - No single existing theory covers all the relevant causal mechanisms. - Some theories assume premises that contradict the empirical record. - Several mechanisms are *implied* by existing accounts but never formally stated. Under these conditions, an **abductive** logic is more honest than a deductive *theory-test*. We reason *from* the systematically organized evidence *to* the best available explanation, acknowledging theoretical uncertainty rather than pretending it away. This is fully compatible with the Bayesian-logical foundation defended by Fairfield & Charman (2022), who explicitly note that Bayesianism "mandates an iterative 'dialogue with the data,' which mirrors how process tracing is usually conducted." --- ## 4. Repository Structure / Estrutura do Repositório ``` process-tracing-database-design/ ├── README.md <- this file ├── NEWS.md <- running log of methodological decisions ├── METHODOLOGY.md <- living synthesis of guiding premises ├── REPOSITORY_PLAN.md <- reorganization plan and design rationale ├── HISTORY.md <- narrative chronology of database evolution ├── process-tracing-database-design.Rproj │ ├── md_documentation_previous_versions/ <- archived versions of documentation files │ ├── README-v0.md (pre-2026-05-26 — original "research data" framing) │ ├── README-v1.md (2026-05-26 morning — first methodological-development framing) │ ├── NEWS-v0.md (2026-05-26 — initial NEWS log) │ ├── METHODOLOGY-v0.md (2026-05-26 — initial premises synthesis) │ └── REPOSITORY_PLAN-v0.md (2026-05-26 — initial organization plan) │ ├── llm_conversations_database-development/ <- AI conversations, references, templates, drafts │ ├── README.md <- folder-level guide │ ├── conversations/ <- LLM conversation exports │ │ ├── rename-suggestions.md │ │ ├── 2026-05-02_perplexity_pt-literature-and-v1-v2-schema.md (Perplexity, 2026-05-02 → 05-05) │ │ ├── 2026-05-03_claude_bayesian-tests_v1-through-v4.md (Claude, 2026-05-03 → 05-24) │ │ ├── 2026-05-05_perplexity_v3-structural-pivot.md (Perplexity, 2026-05-05 → 05-24) │ │ └── 2026-05-24_claude_v5-fairfield-charman-2022-deepread.md (Claude, 2026-05-24 → 05-26) │ ├── references/ <- methodological reading (PDFs) │ │ ├── Fairfield-Charman-2017_Explicit-Bayesian-Analysis.pdf │ │ ├── Fairfield-Charman-2022_Social-Inquiry-Bayesian-Inference.pdf (primary) │ │ ├── Fairfield-Charman-2026_Bayesian-reasoning-replication.pdf │ │ ├── Fairfield-Charman-2026_Appendix.pdf │ │ ├── Fairfield-Garay-2017_Redistribution-Under-the-Right.pdf │ │ └── Fairfield-Garay-2017_Appendix.pdf │ ├── xlsx_templates/ <- successive abstract schemas (renamed from xlxs) │ │ ├── banco_pt_process_tracing.xlsx (v1, 2026-05-05) │ │ ├── banco_pt_process_tracing_v2.xlsx (v2, 2026-05-05) │ │ ├── banco_pt_process_tracing_v3.xlsx (v3, 2026-05-05) │ │ ├── banco_pt_process_tracing_v3_1.xlsx (v3.1, 2026-05-05) │ │ └── v4_banco_process_tracing_...xlsx (v4, 2026-05-24) │ └── drafts/ <- intermediate outputs from conversations │ └── Genereted README_v2...qmd │ ├── Database Process Tracing Reforming Tertiary Education.xlxs.xlsx │ <- populated database (substantive content) └── v7_banco_process_tracing_baesiano_abdutivo_manual.xlsx <- current working database (live test) ``` ### Key distinction: Templates vs. Working Databases - **`llm_conversations_database-development/xlsx_templates/`** holds the *abstract* (case-independent) database schemas. Each version records a methodological decision. These are the intellectual product of this repository. - **`v7_...xlsx`** is a *live test* of the current schema with real data. It exists to stress-test the design, not as a final research deliverable. - **`llm_conversations_database-development/`** stores the AI-assisted conversations (in `conversations/`), the methodological references (in `references/`), the schema versions (in `xlsx_templates/`), and intermediate drafts (in `drafts/`). See its own [`README.md`](llm_conversations_database-development/README.md) for a folder-level guide. --- ## 5. How to Read This Repository / Como Ler Este Repositório 1. Start with [`METHODOLOGY.md`](METHODOLOGY.md) for the guiding epistemological premises. 2. Read [`HISTORY.md`](HISTORY.md) for the narrative of how the database evolved across four LLM conversations from 2 May to 26 May 2026 — the lessons learned, the methodological turning points, what worked and what was rolled back. 3. Read [`NEWS.md`](NEWS.md) as a structured decision log (companion to the HISTORY narrative). 4. Open the most recent file in `llm_conversations_database-development/xlsx_templates/` to see the current abstract schema, or `v7_...xlsx` for the schema applied to real data. 5. For the reasoning behind a specific design decision, check the relevant conversation in `llm_conversations_database-development/conversations/`. 6. For the rationale of *why the repository is organized this way*, see [`REPOSITORY_PLAN.md`](REPOSITORY_PLAN.md). 7. If you disagree with a methodological choice, [`METHODOLOGY.md`](METHODOLOGY.md) §6 on the Van Evera tests shows how tensions are handled transparently. --- ## 6. Conventions / Convenções | Item | Convention | |---|---| | Language | English (academic register); Portuguese translations in codebooks and key explanatory files | | Code | R (tidyverse, readr, dplyr, ggplot2) | | Documents | Quarto (`.qmd`) and Markdown (`.md`) | | Version control | Git + GitHub | | Citations | APA 7 author-date | | Database format | `.xlsx` (manual entry) → R script → `.sqlite` (audit / queries) | | Evidence auditability | Every evidence row must carry `url`, `arquivo_local`, and `pagina_trecho` | | Documentation versioning | Major revisions archived in `md_documentation_previous_versions/` | --- ## 7. Primary Methodological Reference / Referência Metodológica Principal Fairfield, T., & Charman, A. E. (2022). *Social Inquiry and Bayesian Inference: Rethinking Qualitative Research*. Cambridge University Press. Earlier article version: Fairfield, T., & Charman, A. E. (2017). Explicit Bayesian analysis for process tracing: Guidelines, opportunities, and caveats. *Political Analysis*, 25(3), 363–380.