Causal Analysis Workflows

This section of the Tetrad Manual introduces practical workflows for causal discovery and causal modeling.

Rather than treating causal discovery as a single “run an algorithm and read off an answer” step, these pages describe a deliberate, iterative process that combines data exploration, explicit assumptions, systematic search, and model evaluation.

In Tetrad, causal analysis is a scientific workflow, not a black box. This section is intended to support careful reasoning while remaining flexible to different goals and levels of experience.

🧭 What You’ll Learn

This section outlines how causal analysis is commonly carried out in Tetrad, including how to:

Explore your data to form sensible, defensible modeling assumptions.
Decide which classes of causal models are appropriate for the data at hand.
Use Grid Search as a systematic exploration tool for algorithms, tests, scores, and parameters.
Compare candidate models rather than committing to a single run.
Evaluate models using diagnostics, including Markov checking.
Iterate and refine assumptions and searches as evidence accumulates.
Interpret results carefully, including cases where conclusions remain limited or partial.

📌 Why a Workflow Matters

Causal discovery is inherently underdetermined: many different causal models may be compatible with the same data.

Rather than hiding this uncertainty, Tetrad is designed to help you work with it explicitly. A workflow-oriented approach allows you to:

Begin with the data, grounding assumptions in observable structure.
State assumptions clearly, such as causal sufficiency or functional form.
Compare alternatives systematically, instead of relying on a single algorithm run.
Rule out implausible models using diagnostics rather than preference.
Identify robust features that persist across reasonable modeling choices.

This approach tends to produce results that are more interpretable, defensible, and reproducible.

🗺️ How the Workflow Is Organized

The causal analysis workflow in Tetrad is organized around the following pages:

Data Exploration
Inspect datasets and identify features that inform modeling assumptions.
Choosing Assumptions and Methods
See how data properties (e.g., variable types, potential latent variables) guide methodological choices.
Grid Search: Systematic Causal Discovery
Learn how Grid Search supports controlled, comparable exploration of algorithms and parameters.
Model Evaluation and Markov Checking
Use diagnostic tools to assess whether candidate graphs are consistent with the data.
Interpreting Results
Learn how to read causal outputs carefully and communicate conclusions with appropriate caution.
Case Studies
Worked examples demonstrating the full workflow, from data exploration through interpretation.

Although these pages are presented in a logical order, causal analysis is rarely linear. Revisiting earlier steps as new insights emerge is both normal and expected.

🧠 Practical Advice Before You Begin

Treat assumptions as working hypotheses, not fixed truths.
Use visual and exploratory tools early.
Prefer systematic comparison over one-off runs.
Focus on features that persist across models, not just the highest-scoring output.
Keep notes on decisions and revisions; this improves interpretation and reproducibility.

🙌 Where to Start

Begin with Data Exploration to understand your dataset and form initial assumptions before moving on to systematic search using Grid Search.