Studio 5: Develop the pathway
This studio is part of a book in active development and has not yet been through the author’s review. Content may change as the review advances.
What you can defend when you leave. Commit to the research pathway your question and your licence actually support, and say what that pathway can and cannot establish.
The milestone ahead
Five pathways run through this book, and your project takes exactly one as its primary road. This milestone makes the commitment explicit: which pathway, what licenses it, and what it cannot establish no matter how clean the execution. Declaring it now protects you later, because a result can only be defended from inside the pathway that produced it.
This studio closes with Milestone 5: Your pathway, declared, a short chapter of its own after the lessons. What it asks you to produce. Research Contract v1: objective by target and reach by data strategy by warrant, with the pathway declared and its limits written before any result exists.
Milestone 6 puts data and measurement under governance so the declared pathway has something trustworthy to run on. The milestone chapter keeps the working details: the practice steps, the versioned record, and how the artifact is assessed.
Choosing your pathway
Five pathways run through this part of the book, and your question plus what you are allowed to do picks one. Read the row that fits, then read one that does not, so you can say why yours is right. The first four adapt the design library in RDSS (Blair et al. 2023); prediction is this book’s added fifth pathway.
| If your question asks | and you | your pathway is |
|---|---|---|
| how much, how many, or how distributed something is | observe the world as you find it | observational descriptive |
| whether one thing causes another | observe the world as you find it | observational causal |
| how much of a property exists, measured cleanly | assign conditions yourself to measure it | experimental descriptive |
| whether an intervention works | assign conditions yourself | experimental causal |
| what will happen to cases nobody has seen | fit and honestly score a forecast | prediction |
Two cautions. Assigning conditions does not make your question causal: a randomized study can be measuring a property the world already has. And a causal question whose data cannot identify it stays causal — it becomes “causal, currently unidentified”, never quietly descriptive.
If your project combines pieces of several rows, read the complex-designs lesson as an overlay after you have declared your primary pathway.
The lessons in this studio
Read them in order, working each lesson’s It is your turn as you go. Each one builds a piece of Milestone 5; beside each lesson is the piece it hands you.
- Lesson 14 — Observational Descriptive Research: either your four boxes and selection diagram, if this is your pathway, or the classification drill that rules it out.
- Lesson 15 — Observational Causal Research: either your causal diagram with its leverage named and priced, or the drill that shows why this road is not yours.
- Lesson 16 — Experimental Descriptive Research: either your revealed-characteristic estimand with its demand and instrument threats named, or the drill run on your own question.
- Lesson 17 — Prediction and Generalization: either your signed four-part forecasting contract with its boundary, or the contrast drill: the plain rejection line, the prediction-time leakage check run on your own features, and two sentences on why prediction answers a different question.
- Lesson 18 — Experimental Causal Research: either your treatment, outcome, and potential outcomes with the nearest threat named, or the drill plus the confounder that blocks the causal reading.
- Lesson 19 — Hybrid and Complex Designs: if your design has stages: the stage-by-stage map, its weakest link, and the one-sentence justification for keeping or cutting it.
When the last lesson closes, the milestone chapter is next. Nothing you write in a lesson is busywork: every piece above is on the milestone’s checklist.